Spent about 2 years improving printing and scanning stack of Linux: CUPS, SANE, AirSane, as well as some legacy drivers, and also x86 proprietary driver emulation on ARM with Box86.
Even that "modern" printing stack in Linux is 20+ years old, there's still such an unbelievable amount of basic bugs and low-hanging-fruit optimizations, that it's kinda sad.
Not to mention that it still maintains ALL its legacy compatibility, as in supporting ≈5 different driver architectures, 4 user-selectable rasterizers (each with its own bugs and quirks).
The whole printing stack is supported by 4 people, 2 of whom are doing that since the inception of CUPS in 1999. Scanning is maintained by a single person.
Ubuntu 26.04 LTS is expected to be the last version with CUPS v2. CUPS v3 drops current printer driver architecture and introduces proper modern driverless printing with the wrapper for older drivers. Many open-source drivers are already use this wrapper, but expect a huge disarrangement from the users, as none of the proprietary drivers would work out of the box anymore.
Nah, I got into printing because nobody made a commercially available print server, and I ended up making my own, with all the involvement in the stack in the process.
I wish Openprinter luck, as it has been announced in the end of September but nothing out there yet, not even the crowdfunding campaign.
Great initiative. I wonder, how likely is it for a complete beginner to break their own printer or scanner by making a mistake in driver implementation? Or is it possible to work on hardware support without having a physical device? I assume it is impossible to test each and every one printer and scanner, so there is probably some clever tricks there, right?
I work mostly with the old microcontroller-based cheap consumer ("GDI") USB models circa 2000-2010, these are hardly possible to brick with software, as some of them even don't have a firmware and expect the PC to upload it on each power on sequence.
The hardware safety mechanisms are usually robust (USB communication is handled by "Formatter Board", all the mechanical stuff is in the "Engine Controller" power).
Newer Linux-based models have filesystems, software, and vulnerabilities, printer hacking on Pwn2Own is an every year common occurrence. These could be permanently bricked by the software in a common sense, and would require a firmware reflash using the bootloader or external means.
>Or is it possible to work on hardware support without having a physical device?
Absolutely, but for me this is very inconvenient—like the debugging over the phone.
>I assume it is impossible to test each and every one printer and scanner, so there is probably some clever tricks there, right?
Not much, unfortunately. There's ongoing work on modern (driverless) printer behavior emulation, but it is under heavy development and not ready yet: https://github.com/OpenPrinting/go-mfp
Nothing for the older printers and scanners which require it's own driver, of what I'm aware.
> Thank you. I have thrown printers out the literal window.
I have literally thrown one of those "winmodems" [1] out of the window back in the days. I then went out and drove with my car on it. I then put it in a bench vise until its PCB shattered to pieces. Utter destruction, much to the amusement of my brothers.
These were the days.
And big thanks to GP for his work of CUPS / Linux printing.
Over the holidays I was building a wooden birdhouse with a Unifi Protect camera and a small web interface that automatically identifies the birds and shows me a simple overview over which type of bird visited how many times.
Thanks, it's named after a similar commercial project called Birdbuddy but I didn't want to pay for a subscription and potentially have some e-waste on my hands in a few years. I also wanted to have it integrated in my existing Unifi setup.
Ah yet another glorious yak shave... "Hacker refuses to pay for coercive subscription service which isn't even compatible with the home setup, and spends inordinate time and resources making the thing themselves.".
Random assortment of projects as time allows with the $JOB.
- Prototyping a cute little SSH-based sorta-BBS, inspired by the Spring '83 protocol, but terminal-centric rather than web-based. It's called Winter '78, and if we get another Great Blizzard this year, I'll be able to make some progress on it!
- Another prototype, for an experimental HPC-ish batch system. Using distributed Erlang for the control plane, and doing a lot of the heavy lifting with systemd transient units. Very much inspired by HTCondor as well as Joyent's (RIP to a real one) Manta.
My meta side project for building other side projects: https://bodge.app/
I've always had a bunch of small side projects that I want to do that aren't worth the overhead required to actually put them together & keep them maintained. So, I built a small Lua-based FaaS platform to make each individual project less work whenever inspiration strikes. So far I've built:
> My meta side project for building other side projects
Looks cool, congrats on putting it out there as priced service!
And, same!
Except, it's just a repo organisation system (structure, conventions, and tools) that lets me share common "parts" across multiple "projects". No monolithic frameworks here.
Libraries are functions. Apps are objects.
However, normally, we use these as distinct artefacts, eventually leading to the "diamond dependency" problem (and lots of other annoying development-time stuff caused by libs / code that is "over there" (elsewhere)).
My "meta side project" solves, essentially the Expression Problem as it manifests in source code management (particularly, cross library / service / project feature development).
Thanks for sharing! I signed up and tried it for something simple (storing a message via POST and displaying it in HTML via GET) and it was delightfully easy & approachable.
Great to hear! And thanks for saying so. I've definitely tried to make it as simple and straight forward as possible, but I really didn't know if it would be simple and straight forward to anyone but me.
This is great. One thing that is not immediately obvious to me is what kind of authentication it supports. Needing to authenticate the caller is the #1 piece of boilerplate that gets in the way of me completing mini-projects. It’s not hard, but it’s definitely a nonzero amount of effort. And this is coming from someone who has implemented many, many auth flows professionally.
It doesn't say because there's no special support for any auth protocols. Long-term I want to have out-of-the-box support for things like OAuth (for user-facing auth) or mutual TLS (for device/service auth). _Technically_ there's currently support for the cryptographic primitives required to do JWT (I added that because I wanted to support WebPush w/ payloads for myself), but those aren't documented because I intend to remove the current slightly-hacky custom APIs and replace them with some off the shelf libraries, but I'm still figuring out user-added libraries (and on top of that I'll also need to figure out support for native libraries).
Are there any auth protocols / flows you think would be important to support?
> Are there any auth protocols / flows you think would be important to support?
- I think API key passed via basic HTTP auth would get you pretty far. This is ideal for serving machine-machine requests and just requires that both parties can securely store the secret.
- OIDC is great for interactions that happen in the browser or if the function is serving multiple users, but is more complicated to setup and/or use correctly.
OpenID connect is probably the best for contexts where you want something served by multiple users and those users are humans.
> _Technically_ there's currently support for the cryptographic primitives required to do JWT (I added that because I wanted to support WebPush w/ payloads for myself)
This is probably a good intermediate solution FWIW - expose signature verification and HMAC APIs and allow a user to bring in their own implementation.
I know the homepage needs way more answer to "WTH is it?", I just don't really enjoy doing the 'marketing' side of things. I hadn't really considered just throwing something informal up there, but I guess I don't really know _why_, so, thanks for the suggestion.
After years of wondering what the post-UNIX paradigm of computing could look like, these past few months I've been prototyping a software platform built around capabilities and message passing, targeting Linux and bare-metal RISC-V. The big ideas I'm pursuing is a stackless design running on a flat address space with lightweight processes to minimize message-passing latency, and all the benefits of capabilities so every process is sandboxed and has only access to capabilities they have been explicitly passed via message passing.
I've also built a RISC-V emulator to integrate with this platform, so eventually it'll be able to run native binaries written in any language, completely sandboxed, completely built around message-passing. Basically a native, low-level BEAM-like platform to build an entire operating system and user-space.
While my day job is writing boring applications, this is the stuff that keeps me awake at night, and I would love so much to talk and write more about this, about the trial-and-errors I'm facing, but it's still so much in flux every week I'm exploring a new approach. Most of my work has been around the stackless scheduler, and I have a plan to achieve preemption for long-running or misbehaved tasks without having to compromise on memory usage (i.e. without giving each process its own stack and allocate memory for context switching).
Eventually I'd like to layer on top either Cap'n Proto or another high-performance serialisation system to create a distributed, introspectable environment of object-capabilities that are sending typed messages between each other, achieving the ultimate goal of creating an unholy hybrid between Smalltalk and the Erlang VM.
God, how I wish I was paid to work on this type of problems :-)
If this sounds close to your area of interests, please send me an email and I’d love to chat.
Script to auto-rename screenshots with Claude Code
I'm sure there are a ton of other projects out there that do this, but I couldn't find one that fit my needs exactly, so I threw this together in a few hours.
claude-image-renamer uses Claude Code CLI to analyze screenshots and rename them to something actually usable. It combines OCR text extraction with Claude's vision capabilities, so instead of "Screenshot 2025-12-29 at 10.03.10 PM.png" you get something like "vscode_python_debug_settings.png".
A few things it does:
Handles those annoying macOS screenshot filenames with weird Unicode characters
Uses OCR to give Claude more context for better naming
Keeps filenames clean (lowercase, underscores, max 64 chars)
Handles naming conflicts automatically
If you're on macOS, you can also set this up as a Folder Action so screenshots get renamed automatically when they are saved to a folder, typically ~/Desktop. This is useful if you take a lot of screenshots and hate digging through "Screenshot 2025-12..." files later.
I bet you are running some personal server or tinkering VM.
Why not use a dedicated, coherent tool like Immich to manage all your image organisation needs? I don't see how this renaming would help me in any way, to make use of old screenshots left in random places.
Maybe instead of shooting people down in “What are you working on” threads (I see you have priors), you can instead share something cool that you’re working on.
It’s an open source project that basically turns your kubernetes into a developer friendly PaaS.
Just crossed 2k apps on the cloud version, no idea how many people run it locally, and thanks to a generous sponsorship from the Portainer folks, I’m able to work on it close to full time.
Over the last two years I spent a surprising amount of time analyzing XSD files to understand regulatory changes in financial reporting. It was repetitive, painful, and unnecessarily manual.
In the past two months I built https://xsdviewer.com
to make working with XML Schemas simpler: visual structure, navigation, diffing, and faster understanding of what actually changed between versions.
Right now I am iterating on new features and performance improvements. If you regularly work with XSDs or XML based standards and have ideas, pain points, or feature requests, I would love to hear them.
I’m genetically engineering yeasts to make subtly flavored breads. I’ve already done grape aroma, now working on wintergreen.
Also working on a red chamomile (using beat red biosynthesis). Just for fun. Red chamomile tea!
The idea is to have niche invite-only genetically engineered flavors that I can bring to parties around SF :) what’s more special than a genetically engineered organism that you can ONLY get if I’m there? Good calling card
This is so cool. Also, it sounds like a cheeky plot to a zombie apocalypse or global contagion movie.
How subtle are the flavors? Unsubtle enough that an oblivious taster might ask, "Does this bread taste like grapes to anyone else here?" Or does one need guidance to search for the flavor?
I really need to do a write-up. I kinda just whip up the easiest path and do it.
For example for the grape, I needed to knock out some tryptophan synthesis genes so I could redirect the bioflux. Problem is that in bakers yeast they have a whole buncha copies of their chromosomes, so I had to knock out one of the genes and replace it with a different gene from grapes. Did that with a quick lil CRISPR switch.
Had to electroporate tho because the transformation rates on wild/bakers/non-lab yeast are so garbage
How much expensive lab equipment do you need to do something like this?
"A quick lil CRISPR switch" sounds like "oh just my homemade fusion reactor hooked up to my kitchen warp drive" to me, yet you make it sound so simple!
Built a parking garage occupancy tracker for my city (https://www.parkeergaragesdelft.nl/) after a city council decision on car-free city center policies got delayed due to missing data. The city only publishes real-time availability with no historical record, so I started logging it. The site now shows historical data plus analysis of weekly patterns. (Note: website is in Dutch, but the charts should be self-explanatory.)
Now working on a second tool that monitors public reports on illegal dumping, broken streetlights and more. It tracks how long the municipality takes to resolve them
Lately, I've developed an interest in local politics and started reading policy documents, following city council meetings and even lobbied for a local park. Presenting public data clearly can help shape opinions and keeps the city council accountable, especially important with the upcoming city council elections.
Since the last time I posted on HN it's gained a decent amount of traffic and users. I'm particularly happy with the jobs section, which is growing into a high signal-noise source for European tech jobs: https://techposts.eu/jobs
The reason I started this website is because so much incredible innovation and growth in Europe flies under the radar. If you ask Americans many will say it's just "banks and museums", stagnant, or worse. But the reality is there is a huge spectrum of exciting companies starting and growing here. We have space launch companies, battery companies, AI companies, and a whole bunch of other interesting stuff. It's an exciting time to be a European in tech!
"Ask TechPosts.eu: Is there an EU Cloudflare alternative? I'm interested in CDN, DDoS mitigation and basic web security." Links to https://techposts.eu/post/153
For a couple of years now, I’ve wanted to build a service that lets you subscribe to new version notifications for any kind of software—desktop apps, drivers, packages, and more. As a Unity developer and consultant myself, I’ve always wanted to know the moment a new version becomes available. I value being at the bleeding edge, especially since a large part of the value I bring as a consultant comes from both the breadth and depth of my knowledge of the Unity engine.
That’s why I’m currently working on a niche version of the broader VersionAlert idea specifically for Unity, which is why the domain currently redirects to https://versionalert.com/unity
.
The only other service I’ve found that really addresses this need is https://newreleases.io
. I actually spoke with the very nice husband-and-wife team behind it and even offered to buy the whole project, but their asking price was about 10× higher than what I was offering—which I understand, no hard feelings. I still wish them the best of luck.
If anyone is aware of other projects or services in this space, I'd be happy to learn about them and chat.
Built my own Spotify recommendation egnine after getting tired of Spotify’s repetitive recommendations.
You get to choose the genres you're interested in, and it creates playlists from the music in your library. They get updated every day - think a better, curated by you version of the Daily Mixes. You can add some advanced filters as well, if you really want to customise what music you'll get.
It works best if you follow a good amount of artists. Optionally you can get recommendations from artists that belong to playlists you follow or you've created - if you don't follow much or any artists, then you should enable that in order for the service to be useful.
Working on WordPecker, trying to create something like a personalized Duolingo but for vocabulary you actually encounter.
Started when I was struggling to read books in English. Pushed an open source version back then (https://github.com/baturyilmaz/wordpecker-app), later added more features, and now working on a mobile app.
My end goal is to build: an AI language learning companion that knows what you read, listen, and watch, knows you as a friend (real life, who you are), then helps you improve using that context. If you're B1 at language, it creates a personalized path to get you to B2, then C1, and so forth using your context.
That's what good lights go for, they're niche and people pay good money for them. I really liked my office lamps and I wanted to get the same for home, after a quick search I discovered they were 1300+ euros
I'm working on a native MacOS Postgres client: http://github.com/ravelantunes/Searchlight
I've been slowly chipping at it over the last couple of years as I find time, and I like it enough to already use it as my primary client (although still buggy here and there).
I am working on a local RAG LLM designed for lower end PC's - ability for people to try out searching their own documents, seeing it was such a learning curve to get to this stage - hoping others can learn from my mistakes.
I've been working on a music-making website since late spring. This is my first real frontend project, and I'm writing it in pure Rust using Leptos (so far, haven't had to write a single line of JavaScript!)
Most of my work so far has been on the actual music-making interface, but I'm beginning work on the backend now. I've only worked with Django before (for a school project at Georgia Tech), so I'll be deep in the `sqlx` documentation for a while.
There's no manual, so use at your own risk (it's similar to tracker programs like FastTracker and OpenMPT): https://mondobe.com/tracker
Fellow yellow jacket here!
This is great start, I'd love to see mondobe evolve into something like cursor for music production. I've done music making on DAWs like fl studio and i wish if we could reduce some friction there for users
I've been working on a voxel engine in my spare time for 10 years, this year. I wrote everything from scratch, from the memory allocators, font rasterizer and a programing language all the way to collision detection, renderer and gameplay code. It's been a journey!
Building https://lenzy.ai - helping products built around chat with AI (think Lovable or Cursor) reduce churn and prioritize product improvements by analyzing their user's chats.
I started about 3 months ago, focusing on making my 2 early adopters happy. One of them is ready to start paying soon!
Working on https://chessnawk.vercel.app/ – a set of auto-grading tasks for middle school students learning programming.
Comes with Robot, Turtle, HTML/CSS (pixel diff tasks) and a gradual introduction into programming concepts. Currently on JS, literally right now GPT-5.2 helps me adding Python.
I've integrated a simplified clone of Replicube. I hope to integrate ideas from Human Resource Machine and Turing Complete later this year.
I bear heritage of Eastern Bloc-typa math/programming olympiads, combined with front-end/product skills. So I kinda owe to ship this thing to community of fellow secondary school CS teachers.
Surprisingly, nothing comes closer in terms of depth and usability in the classroom for ordinary 12yo kids.
I test this thing 5 times a week in my classroom and I constantly polish it at night. 100% vibe-coding.
I've been experimenting with using LLMs for a content recommender system. Specifically I've built a news reader app that fetches news articles from multiple RSS feeds, uses an LLM to first deduplicate and then score them. The user can then rate the articles and those ratings are used as few-shot examples in the LLM scoring prompt. Any resulting low score articles (uninteresting to the user) are hidden by default and visible ones scaled by their score on a dynamic CSS grid like on a traditional newspaper front page. Looking good so far, but still testing and tweaking.
One thing I’ve always disliked about RSS (and this could actually fix it) is duplicates. When a new LLM model drops for example there are like ~5 blogs about it in my RSS feed saying basically the same thing, and I really only need to read one. Maybe you could collapse similar articles by topic?
Also, would be nice to let users provide a list of feed URLs as a variable instead of hardcoding them.
Looking at the console messages with the LLM reasoning, it does seem to work quite nicely for deduplication. Your example is probably even a lot easier than news articles, where you can have many articles from different viewpoints about the same event.
I don't actually plan to run this as a service so there's some things hard-coded and the setup is a bit difficult as you need an API key and a proxy. Currently it's just experimentation, although if it works well, I'll probably use it personally.
Ive been working on something similar, have you had any issues with the LLM not giving you back a full response for all the input? Ive been using Chat gpt but even on the same request sometimes id give it 20 things to rank, and Id just get back 3 results
No, it's been working without problems so far. I'm using Anthropic for what it's worth. I ask the LLM to first do some reasoning and then return a JSON array on the final line. Sometimes I've seen some Markdown backticks there, but no irregularities more than that.
Previously, I made a live win probability model for the 5v5 arcade game Killer Queen Arcade from their game events API.
Now I am trying to use that model to make:
1. A post game instant replay that shows the most important/pivotal moments from the most recently finished game. Some arcades have a seperate display for observers, it could work well there, or as good filler between matches on twitch streams.
2. A personalized per tournament/yearly highlights recap.
If it works well, it might be a kind of tool that generalizes well for summarizing long twitch streams for Youtube.
I'm working on a CUDA implementation of Forman-Ricci curvature-based clustering (I checked online and saw that there is currently no GPU/CUDA implementation, so I thought why not do it). Hopefully it would help with my CV.
A friend and I noticed recently that the last remaining pieces of the puzzle for a full line of "trustworthy technology" have recently dropped into our laps while we weren't looking.
What are those? You know… open, freedom, and privacy respecting technology. Recent products are the Starlite tablet, Furilabs and Fairphone. We've been waiting for these products, well over fifteen years since the introduction of the iPhone and iPad.
Despite posting many times, we haven't been able to start a discussion here. Maybe we don't know the right key words, or posted too late in the day, not sure. But I know someone is interested in the subject because there have been three huge discussions this week about how "Linux is good enough now."
That has been true for a decade in my experience, but no one seems to be talking about the new mobile hardware available. I hope to work on bringing these efforts together.
Not as impressive, but - making my website weirder. Not a frontend dev at work.
Top of my ideas now: add "ask your LLM" buttons to my email submission form that opens ChatGPT/Perp/Claude and auto-fills a query asking it why you should be friends with me.
Sample links (I hope it says nice things about me!):
buddy, don't be so modest, this is an ambitious project and I hope you succeed ! This is tough work, I tried to build a Webflow-ish editor myself and it nearly destroyed my sanity.
Free to play FPV racing game with a global leaderboard that can run on integrated grapics in the web browser. Focus is on realistic physics (aero, PIDs and filtering, propwash, signal noise).
Doors - A first person, exploration game/experience that I built from scratch.
Doors lets you explore URL addressable 3D rooms that link together seamlessly via portals. The idea is that people would upload rooms to the internet (to github, S3, whatever) and connect them together to form one giant inter-connected space that would be a real trip to explore.
Right now rooms consistent of a:
- Manifest JSON file that points to requisite resources and configures portals
- An optional skybox
- An optional background music track
- A .vox file containing voxel terrain data
Portals can be arbitrarily sized and everything is prefetched/loaded seamlessly in the background.
I'm nearly done - I just need to add in a very lightweight interface and give the code a bit of a spit shine (I will open source it - so I want it to look pretty)
EDIT: As an aside, I finally decided to give this whole Claude Code thing a go - I purchased a max subscription and I'm trying to write as little code as possible. I certainly wouldn't call what I'm doing "vibe-coding". I discuss a feature in plan mode (incl. how I want to implement it in high level terms) iterate on the plan 2-3 times until I'm satisfied and then let it rip. I'm both very impressed and quite frightened by the productivity boost...
Working on two topics that I like : browser multiplayer games and experiences, and tools for developers (mostly used in my side projects).
First game is https://geocentric.top where users can sculpt a planet collaboratively and place trees or houses (for now very limited). I plan on making it an idle sim where players will be able to interact by dropping some food/events for the creatures on it to evolve through time.
Second, a remote logger/metrics/user management tool where once can track all their logs, live metrics, uptimes, identify users, etc. I hope to have a v1 during this first quarter and I'm currently my first user as I have it hosted at https://app.getboringmetrics.com to centralize all my side projects into a single platform.
This is highly polished indeed and I'm likely going to try it, mostly for its animations regarding dropping plants and rocks as I lack micro animations for it in my game.
Thanks! This is also what I love doing as realtime games are something like 90% of my side project. This was a real fun prototype to create as everything regarding terrain is done through shaders (even user cursors). Will soon add a way for people to plant "flags" with small texts that you can highlight.
I’m building a Java stack for the x402 protocol. I believe today’s payment systems are unnecessarily broken and overly complex. I really like the idea of a protocol that could simplify this space while remaining truly open and not “owned” by anyone—meaning anyone can use it without needing authorization from the protocol’s creator.
I just came across x402 recently and am excited to build upon it. My current project[1] will have a tipping feature and it'd be neat to give people the option to not use fiat.
I built a browser plugin called "Visionary" that overlays meaningful descriptions and context directly onto stunning pictures of the day.
I noticed that existing picture-of-the-day plugins were built over two decades ago and never evolved to harness the capabilities of modern artificial intelligence. AI can transform the picture viewing experience by distilling complex descriptions into accessible insights and providing references to explore the core concepts in the photo more deeply.
I’m working on a real-time tracker for the Canadian Parliament.
While the official data is technically public, it's practically inaccessible (buried in XML feeds and legacy sites).
Phase 1 was building a modern ingestion engine and freeing the information to make it more accessible. The goal is to make legislative data as accessible as sports stats.
I'm almost ready to launch the MVP; I'm just doing some bugfixes and testing the database now! (If you want an early look at the MVP, my email is in my profile.)
The next phase is what I'm most excited about: visualizing this data and using LLMs to provide insights.
This is a really cool idea! I remember once looking at all of the actual concrete actions taken and being surprised at what was actually done (not very much at the time!). Maybe it is better now.
I've been thinking about something similar for my city council. Would be interested to compare notes! Your email seems to be missing in your profile, but you can find my email in mine if you want to chat.
I'm on a deep dive fine-tuning how I organize and manage my personal knowledge base - focused on entity extraction and strategic information retrieval and based on the AgREE paper from Apple[0] and persisting it in Memgraph.
I've got a nice ingest, extract, enrich process going for the graph - I'm currently working on a fork of claude-mem[1] that uses the graph as a contextual backend for agentic coding workflows.
Thanks! Is it that you can't see the graph view at all? It uses DeckGL which might be a bit too fancy and may not work on all browsers possibly. It works on Chrome/Safari on my iPhone and Chrome on Arch.
I didn't realize that you actually could provide a working link back to Hacker News but it seems HN does support that. Thanks, I will give that a try!
Oh ok. Yeah, it is just not really complete yet. What this does is displays a UMAP of the embeddings for the selected feature and you can mouse over for labels. It isn't identifying clusters itself yet, finding exemplars, etc. Right now, points that are close to each other are just semantically close - as judged by UMAP.
Personally, been making a low fidelity exalidraw-like calendar app: https://letswalnut.com.
There’s a real-time collaborative workspace-oriented version, too.
Professionally, working on “Magic Draft,” a feature in Ditto to help designers and writers create the “draft and a half” directly in Figma, which uses a hierarchy of all your context (text, Ditto metadata, the design, your style guides, etc) to write really good starting point copy.
AFAICT, the future of software development looks like a lot of unprofessional software development.
Not unlike digital photography and Instagram. Has it killed film-photo divisions of photography companies? Yes. Has it put professional photographers out of business? Hardly, and in fact, the opposite. What the ubiquitous phone camera has done, is expose a lot many more people to the steep challenge of making truly good photographs. It has raised the average population-scale level of photo-erudition and taste to ever-more sophisticated levels. And, it has pushed the envelope on what photography can do.
So---assuming the AI overlords prevail (which I'm deeply skeptical of, but suppose a trillion dollars are right and I'm wrong)---what happens when LLMs allow anyone to vibe-code their own SaaS or Database or IDE or bespoke health-monitoring app or whatever...?
I've been working on an iOS camera app to take natural-looking photos with reduced post-processing. The goal is take photos that look like what you see.
I just updated the RAW pipeline and I'm really happy with how the resulting photos look, plus there's this cool "RAW+ProRAW" capture mode I introduced recently.
I initially released it early last year and have been using it as my main camera app since, but I haven't mentioned it in one of these threads before. Unfortunately this post has come just a bit early for my most recent update to be approved; there's some nice improvements coming.
Curious what motivated you to create this camera app when there are a handful of well established apps that already do unprocessed imaging. What does Unpro do differently?
At the time I created it there weren't any that did the end-to-end process with ProRAW in a way that I liked. And I got really tired of manually editing every photo I took, so I built the app for me.
Plus, having full control over the way photos look, I've customized the output to match my taste; I don't think there's any other camera apps that produce photos quite like mine.
I'm currently working on Codebox, a remote development workspaces provider. After trying many existing solutions, I realized there wasn’t a self-hosted option that was both simple to set up and easy to use, so I decided to build one.
With Codebox, you can define workspace templates or load workspace configurations directly from Git repositories. At the moment, it supports Dev Containers and Docker Compose, with plans to expand to additional configuration methods in the future (for example, Terraform).
Below are some resources where you can learn more about the project, including an article that explains how Codebox works and the source repositories (mirrored on both GitHub and GitLab):
I built a free USCIS form-filling tool (no Adobe required) USCIS forms still use XFA PDFs, which don’t let you edit in most browsers. Even with Adobe, fields break, and getting the signature is hard.
So I converted the PDF form into modern, browser-friendly web forms - and kept every field 1:1 with the original. You fill the form, submit it, and get the official USCIS PDF filled.
I built an in-browser data explorer for datasets too large for spreadsheets but too small to justify spinning up SQL. It runs entirely client-side using DuckDB-WASM, no server uploads accounts or anything.
The core idea is a visual DAG where each transformation (filter, join, aggregate, pivot) creates a view node. Nothing materializes until you need it, DuckDB executes the full chain on demand so you can build deep pipelines without copying data at each step.
Input files can be CSV/Parquet/Excel (Excel might not work great).
There's a SQL editor with schema-aware autocomplete, pivot tables with drill-down to underlying rows, and sessions can be exported as files or shareable URLs (the entire pipeline gets encoded in the hash).
Sharing can be granular and you can choose not to embed the files or if files are too big they will not be embedded and the user when opening the link will have to upload the files to restore the session.
The part I find most useful: you can replay pipelines on new data. Share your January analysis, and a colleague runs the same transformations on February's data with schema validation.
Privacy-first since files never leave your browser, it's a static website actually.
I will open source soon, and make it probably MIT licensed.
Also it's a WIP and so it may be buggy (there's not even images on the homepage)
https://repere.ai
I'm in the final stretch of self-publishing my debut sci-fi novel Sector 36. Been working at it for almost 2 years. I'm waiting for the final print proofs to make sure things look good before the "official" 1/31 launch date.
Called "feocask" cuz feo means "ugly" in Spanish and FeO to mean Iron Oxide for Rust. I thought it was funny.
I will admit that I had help from Codex, but I did write most of it myself, and I think the design is coming out kind of neat. I have a very strict "no lock" policy [1], including lockfiles, and this should still be safe to use across any number of threads, at the cost of N^2 reconciliation to the number of threads and a lot more drive space.
I like my design; I have an excuse to use Vector Clocks and Hybrid Logical Clocks and I think it might actually be useful for something some day. I'd like to eventually write something that goes a bit beyond getting parity with bitcask and optionally have the ability to automatically distribute across multiple nodes, but I'm still trying to think of a good design for that, because my current design depends heavily the atomicity of POSIX filesystem commands, and introducing the network introduces latency that would likely greatly degrade performance.
[1] At least no explicit locks. I am using Tokio channels and they are probably using locks in some spots behind the scenes.
Building MatGoat (https://matgoat.com) - management software for Brazilian Jiu-Jitsu and other martial arts' academies.
I train BJJ and kept hearing the same complaints from academy owners regarding attendance tracking, comms, missing payments, etc.
So I'm building a tool for student tracking with belts progression, automated payments, attendance-based promotion criteria, and a tablet check-in system.
Focusing on Spanish-speaking markets first since it's completely underserved. Currently onboarding early academies, and will market it in the US/UK soon.
Over the last 2 or 3 years I've been building a Common LISP implementation in Go so that Go packages can be utilized by LISP code. Building a REPL with lots of interactive features was rewarding as was taking up the challenges of the object systems (CLOS and Flavors) and generics. Just open sourced it at the start of January. https://github.com/ohler55/slip
I think frontier models are getting to the point where we can start to reach higher trust agentic workflows.
As a hardcore AI chat user, I'm often frustrated with the single-agent workflow, where a single context window is used for even very long conversations. If I want to change the topic, open a thread, or go on a tangent, I often end up compromising the main thread and I'm forced to copy context over if I want to dive into something.
To solve this, I'm working on a collaborative AI agent orchestrator that models the solution as a group chat with humans and AI agents, including an agent orchestrator.
You can spawn participating agents with the orchestrator who will decisively route messages to the existing agents, or spawn new agents if needed. Also, you can open agent details and send messages directly to existing agents, similar to threads in slack.
So far, I have MCP integrations working with Linear and GitHub, but plan to add many more.
I've been working on this just over 2 weeks, making heavy use of 4+ concurrent Claude Code agents. This would have been impossible otherwise.
Ooh this sounds good, I feel the same problem but never imagined a solution, having a group chat sounds good, any chance this can be integrated in something like Antigravity?
Nowhere near as impressive as some of the things here but I’ve been slowly growing out a daily puzzle website https://regularly.co - I built the first few games for my wife, no subscriptions, no pay to play, no sign up needed.
Im launching a new word game next week which I’m super excited about. If you do play it and have any feedback do shoot me message!
Ah another nice feature would be blank out game until you hit start and cant reset timer. I like to compete with friends and resetting time makes it so you cant compete.
Got tired of helping enterprises run Concourse themselves, so we productized it. We've deployed and maintained Concourse for Starbucks, Accenture, Sky UK, and others over the years—CentralCI packages that operational knowledge into a SaaS.
Why Concourse over GitHub Actions?
* fly execute lets you test pipelines locally before pushing. No more "commit and pray"
* fly intercept drops you into a running container to debug failures
* Resource-based triggers can monitor anything—not just git pushes
* Full pipeline visualization from dev to prod in one view
* Workers you actually control (no arbitrary cache limits or runner queues)
What we handle:
* Dedicated Concourse instances on high-spec hardware (Ice Lake Xeon, DDR5, NVMe)
* Worker scaling without the Kubernetes complexity
* SOC compliance, auditing, AWS PrivateLink for enterprise
* 24/7 support from people who've been running Concourse in production for years
The pitch is simple: Concourse is the right tool for complex CI/CD, but running it is a pain. We make it not a pain.
I'm working on a complex garden planner to keep track of gardens, orchards, and landscaping projects. Meant for very large and very small scales. Public alpha at the end of this month, I hope.
It is based on a garden designer I made for myself to keep track of my rose garden (I have over 100 roses) and orchard (I have about 15 trees): https://garden.simonsarris.com/
However, my version was very specific to my needs, so this general version requires a lot more work to get it usable for a lot of people.
Continuing to work on Minifeed (https://minifeed.net/), a directory, reader, and search engine for personal blogs. The indexing & searching backend in Typesense, and I'm moving from their paid managed cloud service[0] to a self-hosted VM. It was very easy to start with the reasonably priced managed service, but with the number of feeds/posts growing, I have decided to self-host it. I'm also using Typesense Dashboard, a nice visual tool to do basic administration [1].
Overall, Minifeed keeps chugging along, fetching new posts every day from almost 2k feeds. I'm hoping to find some nice and ethical monetization strategy for it this year.
I needed some pixel art for a different project, so I made 8BitSmith, a simple pixel art & sprite generator.
I'm expecting it is pretty niche, but animations tend to be very time consuming for people like me, and getting quick sprites that I can drop into a platform is a big time saver.
A transformer-based (but not LLM) chess model that plays like a human.
The site right now is very rudimentary - no saving games, reviewing games, etc., just playing.
It uses three models:
* A move model for what move to make
* A clock model for how long to 'think' (inference takes milliseconds, the thinking time is just emulated based on the output of the clock model)
* A winner model that predicts the likelihood of each game outcome (white win / black win / draw). If you've seen eval bars when watching chess games online, this isn't quite the same. It's a percentage based outcome, rather than number of centipawns advantage that the usual eval bars use.
Right now it has a model trained on 1700-1800 rating level games from Lichess. You can turn it up and down past that, but I'm working on training models on a wide variety of other rating ranges.
If you're really into computer chess, this is similar to MAIA, but with some extra models and very slightly higher move prediction accuracy compared to the published results of the MAIA-2 paper
Acme [1], a text editor for programmers. It’s a very unusual tool and I have collected hundreds of notes (as well as dozens of bugs and inconsistencies) from the past two years of using it. Now I feel fully assimilated, I do not take as much notes and it feels the right time to start organizing the knowledge, to put ideas together.
I think that Acme is very underrated in its domain (a tool for experts). The coverage is minuscule. That’s why I am thinking about blog posts, maybe video tutorials. I do not know what should it be exactly.
I do not have any time estimates. I have a very demanding main job (I work as a software developer) and young family that I need to take care of. This project demands focused efforts and selectivity that I can barely satisfy. But I wouldn’t give up, the thing is totally worth it. If it’s going to take years, so be it.
If you have any comments, send me a letter at kalterdev@gmail.com. Have a good day.
Long term memory for dozens of AI tools, designed for power users who want more control and flexibility than native memory systems and who do not want to be locked into any one platform. You can also have the system remember your entire chat history going back years and use this information to help you better in new chats, it sometimes makes chats 10x more useful when I say something like: “Using recall tool, do 10+ calls for 1000 tokens context each to learn about my interests, strengths, curiosities, what I’ve tried in the past, what worked, what didn’t, etc and suggest a new hobby I would enjoy”.
Without long term recall, AI is a super intelligence in your hands that uses the knowledge of the world to give you generic, nearly useless advice because of how generic it is. With long term memory, you have a super-intelligence that knows YOU. This is what MemoryPlugin solves for.
Working on finally releasing the programming language I’ve been working on for quite a while.
I’m setting up the basic site, which is not a huge deal, but I’ve been inspired by more recent language designers having a streaming presence, so I am working through test runs of streaming my development.
I hope to start with demos of the basic language features and then move on to streaming both a reimplementation of my compiler and on a Rocq implementation of the syntax and semantics of the language for proof work.
The language has a rather small niche at first glance, so I’m hoping to use the streaming as a way to explore areas of appeal and maybe draw some interest. A low level concurrent and parallel ‘functional’ language with very non-traditional syntax and a modal, dependent type theory is not going to appeal to everyone, but hopefully I can find some interest eventually, even if just to hang out on chat and talk about the subject.
> A low level concurrent and parallel ‘functional’ language
I'm not particularly familiar with array based languages, but are you inspired by them at all? Seems like a similar concept.
What're your goals for the language? It would be cool to see a parallel execution model unify SIMD, multithreading and gpu. I bet people with a lot of money would be interested if you could apply it to ML
Not particularly inspired by the array languages in general. I do like the idea of a specifics DSL or library that is built from the ground up for ‘vector’ operations (Futhark would certainly be an inspiration here). I do think an (syntax divergences aside) implementation of what is essentially CUDA-like language/library is very possible.
I don’t really have any lofty goals, the language exists to let me have my personal ideal programming language. One of the things I would like to see adopted, by any language, is that the language is able to encode and present provable guarantees about code. So, memory safety, freedom from UB, lack of overflow (integer or stack) is provable in some subset of the language and the proof is done in a logically consistent, verifiable manner.
Of course most of those proofs are not anywhere near feasible for general case code, but the language can restrict allowable constructs in a computation, function, process, or module to a set that makes the desired properties provable by construction.
The language should be capable of unifying and abstracting over SIMD (or other hardware implementations), GPUs, and OS or Userspace multithreading at levels of abstraction from assembly to Haskell or other high level languages.
A specialized programming language for 3D geometry generation + manipulation called Geoscript as well as a Shadertoy-inspired web app for building stuff with it: https://3d.ameo.design/geotoy
There have been lots of cool technical challenges through the whole process of building this, and a very nice variety of different kinds of work.
I'm working towards using the outputs from this language to build out levels and assets for a browser-based game I've been dabbling with over the past few years.
Article about Norway problem in YAML. Draft is ready (link: https://lab174.com/blog/202601-yaml-norway ), beta readers welcome, please let me know if you have any feedback - thanks!
I've created a programming language and game for (my) kids called Stacky Bird. You learn Stack Based programming with Flappy Bird. Inspired by my old HP calculators, the game 2048 and Flappy Bird. In levels you earn a new instruction and at the end you have a complete instruction set, and can solve complicated puzzles. With new instructions you can even go back and solve previous levels more elegantly.
It's not mobile friendly yet, but maybe that'll be a next weekend project. At least you can view a video of a level on mobile.
I’m building AgentOS [1], trying to experiment where agent substrates/sandboxes will head next. It's a deterministic, event-sourced runtime where an “agent world” is replayable from its log, heavy logic runs in sandboxed WASM modules, and every real-world side effect (HTTP, LLM calls, code compilations, etc.) is explicitly capability-gated and recorded as signed receipts. It ensures that upgrades and automations are auditable, reversible, and composable. The fun bit is a small typed control-plane intermediate representation (AIR) that lets the system treat its own schemas/modules/plans/policies as data and evolve via a governed loop (propose > shadow-run > approve > apply), kind of “Lisp machine vibes” but aimed at agents that need reliable self-modification rather than ambient scripts.
I'm working on BacklogAI, an autopilot engineer for SaaS teams.
You connect GitHub, CI, Sentry, and Linear, and it takes tickets all the way to production. Claude writes the changes, BacklogAI handles tests, migrations, feature flags, staged rollouts, etc...
It’s clearing months of backlog work in hours, and a couple teams I’m working with have already stopped hiring because it’s cheaper than adding more developers. It's crystal clear that developers and designers won't be needed in a couple of months because claude increases productivity by at least 120x and one or two PMs can do pretty much everything.
My stack is Claude, v0, nextjs, shadcn, clerk, supabase, vercel.
I'm working on https://fullremote.it, which started as a side project and is now turning into a real business.
It's an "italian-oriented" curated remote-jobs and remote-work community, currently ~10k subscribers across newsletter and Telegram. What began as "let's share good remote roles" is evolving into paid job postings, sponsorships, and coaching for companies and devs.
I'm trying to make a Chinese Hanzi variant of Wordle called Handle 汉兜 [0] available as a Discord activity/app so my Discord groups can have fun solving puzzles just like we've been having with NYT's official Wordle Discord app[1]. A Discord app is just a webview or iframe embedded inside Discord that you can launch.
Over the holidays I built a simple website which lets children (of all ages) easily draw something and then bring the sketch to life using AI and a prompt.
Only shared it via Show HN so far, and am still regularly getting some creative submissions. Will be sharing it at an art festival later this year so kids can have a more active role when visiting.
I'm working on a subscription-based short-form video site called NICKEL[1]. I felt gross about using YouTube but wanted to share my gaming clips, so I made my own thing. Then I thought about making it sustainable so here we are. I'll have an update to the mailing list out in a few hours, I'm "building in public."
My feature-complete deadline is April 15th and I think I'll make it. If you want to check out the UI, visit the explore[2] page. I have it setup to redirect to a public video while I work on the intended UI (a design challenge I've never tried before but we've all seen). I'm thoroughly enjoying figuring out how streaming video works and how best to optimize things.
You know how Spotify (or any other service) will just silently remove tracks from your playlist because they lost the rights?
No notification, a beloved song just disappeared from your life.
Building a webapp to keep track of your playlists and notify you when a song disappears
Support for Youtube playlists please... having to lookup the watch ID on Google after a song is removed to guess what the track was is a highly sucky part of using Youtube for music.
Many of my enterprise customers face the microservice hairball problem. Adoption of microservices has been "too succesful". There may be 200-400 microservices, and ownership may be lacking. The number is typically growing fast, because it is easier to create a new one than use an existing one.
So I started to work on a side project called "Arch Ascent" for addressing these situations. It seems like it is becoming a kind of a architectural governance tool for visualizing and validating software system dependencies.
- Dependency Graph — Syncs projects from SonarQube and visualizes component dependencies using Cytoscape.js, with graph algorithms for detecting cycles (SCCs), clustering (Louvain), and computing coupling metrics
- Visions — Workspaces for exploring architectural scenarios. Each vision can have multiple versions/variations with different layouts while sharing the same definitions
- Layers — Named groupings of components (e.g., "Domain Layer", "Team Ownership", "GitLab Groups") that can be visualized as colored regions on the canvas
- References — Named sets of components defined via Tag expressions, layer membership etc.
- Statements — Architectural intent constraints that can be evaluated against the actual dependency graph, such as existence, containment, exclusion, cardinality etc.
The plan is to also incorporate Grounds, which are Intermediate stable states on the path from the current situation (ground zero) toward a vision. Each ground represents a releasable milestone that moves the architecture closer to the target vision without necessarily fulfilling all its statements. Grounds enable incremental architectural evolution with well-defined checkpoints.
Stack: Django, HTMX, Cytoscape.js, pyparsing for natural language parsing of References and Statements.
SourceMinder: A “code index” tool that finds symbols in a codebase and creates a single table sqlite database for the index. It uses tree-sitter to parse the AST and add the symbols and what they are (function, class, argument, etc) to the db. I currently have it working with TypeScript, C, Go and PHP. I’m working on adding Perl next, after someone requested it here on HN.
I am writing a Context Retrieval and Context Minimization program for Rust code. I am trying to automate my own workflow in how I use LLMs.
The program is gonna do, what I am currently doing by hand, opening files and copying function/method signatures usually, from files all over the place.
The key here is to fetch into the context window only what is needed for one-question/one-answer and no more, hence the Context Minimization. Context fetched is gonna be specified by the programmer, for example sig(function) fetches only the signature, while @function captures the whole body. sig(Struct) is gonna fetch the fields and signatures of all of it's methods. sig(Trait) similarly.
In my view, giving the A.I. more information than needed, only confuses it and accuracy degrades. It is also slower and more expensive but that's a side effect.
The project is in early stages, for the moment it calls ast-grep under the hood, but eventually, if it works as it is supposed to, I plan to move to tree-sitter queries.
If there is a similar project somewhere I would appreciate a pointer to it, but I am not interested in implementations of agents. My program does not give the A.I. a whole view of the codebase, only the necessary points specified by the programmer.
An eBPF-based Wireguard mesh VPN in Rust using Aya.
The idea is to rely on kernel Wireguard, and process packets in kernel space (via eBPF) for maximum performance and minimal CPU overhead. I plan to use egress and ingress TC to “apply” the policy at both sides. XDP is faster, but only works on ingress, which is not sufficient for a mesh VPN imo.
Netbird already exists in this space, so this is mostly a learning exercise, and maybe a reference implementation for those learning eBPF in Rust.
Goals and constraints:
1. Single digit CPU overhead for multi Gbps bandwidth (probably a bit too ambitious, but we’ll see)
2. Linux only
3. No hole punching or complex NAT handling
4. Basic policy language for L3 and L4 traffic (L7 requires punting packets to a userspace proxy)
An alternative front-end / game discovery service / price tracker for GOG's catalog. Mostly manually enhancing the data from the API (90% heuristics, 10% human effort, as the total dataset isn't large enough for it to be worth doing otherwise), and offering a wider range of filters for it all.
I'm grouping all related products together for a more complete overview, and have recently added library import, where I mostly 'solved' the issue of GOG not recognizing that you own certain games if you got them as a freebie or as part of a since-delisted deluxe edition. Just now starting in on incomplete "series" listings, seeing what'd be involved with making them contain all relevant games, and then exposing that.
I posted this earlier but realized that I had pointed it at a different "What are you working on" post. I think this one (david927) is the defacto "official" one.
Recently, I've been very motivated to make one niche crafted service after another. For myself, family and friends. But struggled to find a compelling hosting solution for projects that only has and will have only a single user for years.
I bought the cheapest mac mini M4 on sale, put it in basement and started working on some cli+daemon to help me automate all things around it.
The biggest risk is security, so probably gonna rely a lot on Cloudflare at start.
I have always assumed this is about individual projects, but let me try talking about our team effort. We just launched exe.dev, and have lots of projects. In particular, we are spending a lot of time on file systems. The easy one is copy-on-write VM cloning, which existing software provides for us. More interestingly, I believe the standard cloud approach of putting everything on a NAS by default is wrong. Doing better here is going to require doing unusual things.
I'm building a newsletter called Tech Talks Weekly[1] where my readers get one email per week with all the latest Software Engineering conference talks and podcasts[1] published that week.
In January, I've released a paid tier[2] where my subscribers additionally get:
1. Access to my internal database of all the talks and podcasts since 2020 (+48,000 in total) where they can search, filter, sort, and group by title, conference/podcast, view count, date, and duration.
2. See the list of the most-watched talks over the last 7, 30, 90 days, 6 months, and 12 months based on number of views.
3. Get category-based view of new talks & podcasts by tech stack, language, and domain
(Software Architecture, Backend, Frontend, Full Stack, Data, ML, DevOps, Security, Leadership and every major language & ecosystem)
I am building an analytics calculator/toy that slowly turned into something useful.
It started as curiosity. I wanted to see if I could express business logic as simple choices and let the numbers fall out on their own.
The app is built interactively in Streamlit. I do not sit down with a spec or a backlog. I add one small idea at a time, refresh the page, react to what looks wrong, then adjust. It feels closer to sketching than programming. The interface tells me what the logic should be next.
Underneath there is a growing pile of rules about the business I am in.
I do not write code in the traditional sense.
I have never coded before and are solving my own problem done by one, this sure feels like magic!!
I am working on a very simple browser extension that allows you to take notes right on youtube. It is open source, offline and everything is locally stored.
After watching the movie „The Amateur“ I started to build myself some confidential looking tools just for fun using C and deployed them as CGI programs.
Vim-inspired, as in it's minimal and fast, text editor on the web. No modal-editing yet but, once the current natural input mode is decoupled, clients can bring their own modes/controllers via the extension system.
It's meant to be embeddable and hackable, serving as a building block for custom IDEs as opposed to being IDE-like VSCode. I felt the web IDE space was uninspired with apps built around VScode/Monaco effectively being hosted a VSCode instance with a pre-installed extension and config.json. (Aside: perhaps there's a business opportunity for VSCode-as-a-service where client apps simply bring their own config). I'm dogfooding this library in building an algo trading IDE.
Ships 2kb and smoothly handles 50+ million line files. 1 billion lines with the high-capacity extension. Also, it can function as a TUI or terminal on the web because the core implementation concerns efficiently rendering plaintext in a fixed-width grid layout.
Another itch of my own to scratch but thought I'd see if I can make some side income as well.
It's a web based bookmark manager with extensions for Firefox and Chrome.
You can easily import and export bookmarks so you have all your data whenever you need.
One main thing I really like that I think makes it stand out is the ability to export the contents of a bookmarked page to an epub file to read later on your Kindle or other e-reader device.
Looking for any constructive feedback on this! Thanks.
Still working on a system for allowing Open(Python)SCAD to model cuts for a CNC using G-code, and just added support for 3D Printing via Full-control G-code:
The only problem I have is that it's so effing expensive to run those games that I can't have a good number of games to claim to be any sort of legit benchmark. BUT so far the games that I paid out of pocket and ran are looking good and I think there is merit to this.
also had lots of fun building on top of Cloud Flare and solving some distributed systems problems while building this.
if you can help me run more games (for science!!) let me know!
I'm branching out and trying something ridiculously different than anything else I've ever done before, I'm gonna try to create a video game that plays on people's sexual attractions, and tries to distract them while the player tries to complete a series of challenges/activities. It'll be like a little laboratory experiment, with NSFW graphics and the player having to remain focused while other things try to distract them.
I've never created a game before, less so a NSFW one, and I'm not sure how it's gonna go, but it is very different compared to other things I've done before. The game itself is done in Rust, compiles to WASM to be run on the web, and I've found 3 artists and one voice actress who is helping me with the art/audio stuff. So far a lot of fun, although managing a fleet (4) of contractors is less fun, although still new so a little bit of fun :)
I've been working on some electronics stuff more recently, and I designed custom PCBs which read the voltage and internal resistance of the 12V NiMH batteries used by robots in FIRST Tech Challenge. It's not always easy to know how charged or healthy the batteries are for a match, or when it's time to get rid of them. They're pretty small and are easy to assemble. Rather that having a screen (which would take precious flash and pins on the ATTiny412), it just uses a WS2812B RGB LED.
I wrote the firmware in Arduino, which was a great learning experience because I typically work with CircuitPython or Go, where I'm less constrained.
A suite of productivity tools for Claude Code, Codex CLI and similar CLI agents. These solve pain points for me, and I build and share them in the repo[1].
I’ll highlight a couple:
- an “aichat” command group that enables continuing work from a session that is at full context usage, by creating a new session and injecting session lineage pointers so the agent/sub-agent can recover arbitrary full details from ancestor sessions. So no more compacting needed.
- aichat search command: TUI for rust/tantivy-powered full text search across Claude and Codex sessions. CLI/json mode for agents to search for past work.
- Tmux-cli tool + skill to enable CLI agents to interact with scripts (including other agents) running in other Tmux panes. Like Playwright for the terminal. Multiple CLI agents can collaborate/consult etc. Agent can run and interact with interactive CLI scripts.
Working on CiteLLM, an API that extracts structured data from PDFs and returns citations for each field (page + coordinates + source snippet + confidence).
Instead of blindly trusting the LLM, you can verify every value by linking it back to its exact location in the original PDF.
My first mobile app! It encodes and decodes SSTV images, allowing you to send pictures over sound waves. Mostly used by amateur radio operators... even though the ISS sometimes also broadcasts SSTV images!
I’m working on a small iOS app for streaming my own media directly from a NAS.
I built it because I wanted access to my music (and videos) on my phone without running a full media server like Plex or Jellyfin, and without syncing files locally. The app streams files directly over a VPN (WireGuard / Tailscale), mainly via SFTP/FTP, and plays them as-is without re-encoding or server-side indexing.
You can browse folders directly, or let it build a lightweight local index for faster artist/album browsing on larger libraries. It started as a personal tool, but I’ve been polishing it after some feedback and opened a small TestFlight beta:
https://helmtk.dev is a toolkit for helm chart maintainers, including a structured template language than can compile into helm templates, and a test suite tool for writing tests in javascript. Super handy I think.
https://blog.atlas9.design is about building a better software experience by solving more of the common stuff from the start: IAM, builds, API design, etc. I'm currently designing and building a Go-based framework to start.
Was really impressed with Claude ability to port models from cuda PyTorch to MLX, so that’s what I have been doing last three weeks; basically I have Mac to test ported models and Gradient vm with PyTorch and Nvidia gpu, Claude can run code in vm, investigate layers and deep analyze the model mechanics and then reimplement them on local Mac with MLX. So far completed port of various models mainly in audio domain, achieving pretty substantial speeds for inference on my machine. Models ported with numerical parity to originals: facebooks omnillingual ASR, Sam-audio, Nvidia sortformer. Planning on releasing this in repo soon.
I'm building an agent builder and execution platform. It's a multi model, multi agent system with lots of tools (bash, databases, APIs, crawler...).
Agents are codified for specific goals. Any business process that needs agent based assistance is broken into workflows and steps. Each step is assigned to an agent. Integrations (API or file access) is requested. Then user can try out, tweak and finally deeply.
The aim is to build a diy platform work configuration, tracing and evals in one place. Code generation is used internally. User doesn't need to write any code.
I'm working on Daestro[0] - a job orchestrator that can directly integrate with cloud providers like AWS, Vultr, DigitalOcean and Linode to create instances for jobs to run on and destroy when done.
Currently I'm working on following features:
- Multi user support (Team) on project level
- Then I'll look into whether to add support for OIDC/SSO now or not
- Alert on slack
- Webhook support
For the past month I’ve been working on a creative / VFX / 3D tool that connects Apple devices into an all-in one node editor: https://subjectivedesigner.com/
With it, you can build interactive experiences, connect device sensors, compose shaders with AI models, orchestrate real-time data flows, and create projects that span across the entire Apple ecosystem.
I’m posting about it regularly on social media and you can see some of it here: https://x.com/sxpstudio
(Though it’s still early and most content is on socials thus far).
It’s done fully in SwiftUI + metal and also a good occasion to ramp up on agentic-powered software engineering. So far it’s been a lot of fun and working really great for me. And to be clear I’m absolutely not talking about vibe coding :-)
I bought a 300 litre aquarium about 6 weeks ago for the living room. Added soil, plants and water. Have spent the last 6 weeks watching plants grow, and snails that smuggled in on plants, multiply. I over fertilised it and left the light on accidentally for a couple of days whilst I was away and experienced an algae bloom, which was interesting. Added some cardinal tetras and amano shrimp yesterday and have spent a lot of time just watching them potter around. Has been a nice change from looking at a screen.
I'm working on https://cloudventory.io – AWS inventory search for teams that don't have the budget for high-end cloud management platforms.
Cross-account, cross-region search. Need to find an IP? Easy peasy. Need to find all the cruft Todd left behind? Search "todd" and see every todd-test-server-1 and todd-alb hanging around.
I've added insights for security, ops, and cost savings – minimal right now but expanding.
A little cool tech detail, I didn't want a backend or to store the information of people's reports anywhere. To get around the requirement, I made the form deterministic and populated on load from a data stream of bits (form is made from booleans or bitfields) from a decoded base-64 query parameter. A cool side-effect of this approach is I can update the query parameter in the URL in real-time as you fill out the form so if you reload it remembers your form without any local storage or cookie use!
I’ve been putting together a rather silly slide-puzzle game for MENSA-level players: things like arranging the tiles by chemical element name according to atomic number, arranging countries by surface area, etc.
It’s absurd and will probably appeal only to the descendants of Ken Jennings.
A beginner-friendly programming language for 2D games where multiplayer is automatic. Intended to be an engaging way for teenagers to learn to code by making games they can play with their friends. Like a blend of Scratch and Roblox. I've been working on this for 3 years!
Trying to visualise magnetic fields of a floppy disk.
So far I can see magnetic fields on a magstripe https://www.youtube.com/watch?v=c8nM4Z-hkTw with two polarisers (one of which I rotate in the video, which is contained in a 3D printed holder with gears I made).
I'm awaiting some different polariser film, to see if I can get it to work with a floppy disk.
Trippy! Please do a large-format visual art installation of the results of this... like, project it on a wall or on some other public surface that is... unexpected.
A system-agnostic language for magic spells with a compiler capable of producing the magic wand movements, incantation, hand signs or magic circle required to perform the spell
I’m working on a proxy MCP server that lazy loads tools to save tokens https://github.com/mquan/nimble. It includes a dashboard for connecting and configuring MCP servers.
I'm tinkering around building "JARVIS" (I didn't want to come up with a clever self deprecating name - this works) - a personal project to manage my life. Integrates into Google Mail, Google Calendar, Trello, GroupMe, EveryDollar. Basically it nags me to do grown up thing and is a better UX than Google Calendar/Trello - I just talk to it and ask it things.
Also experimenting with a new Claude-Code flow; give the bot its own AWS account, Put a bunch of tickets on my personal JIRA, be persnickity about what constitutes "pass" and tell the bot "follow these instructions, pull down tickets until there are no more. Your branch cannot merge until you have integration tests passing in your own dev env first" (I use AWS CDK). Then let it loose to build. The instant feedback loop that Claude has with Build-Code->Deploy to AWS->Run Integration Tests->Address Failures is really nifty fwiw...
I'm working on a continuous chain-of-thought reasoning architecture for generative AI models.
It is similar to Meta's COCONUT. However, instead of the training forward and backwards passes of the reasoning tokens being done serially (slow), they are done in parallel (fast). At inference the reasoning tokens are still decoded serially. The trick is that even though training was done in parallel, the reasoning tokens are still causal and you still get the benefits of increased computational circuit depth.
The kicker is that the architecture is modality agnostic (it doesn't rely on language for its chains of thought), and I want to use it to bring COT reasoning to protein and anti-body generation. Basically, I hope for it to be the OpenAI o1 or DeepSeek R1 for domain-specialized scientific AI.
I’m building a quantum photonics experiment that is a variation of the quantum eraser.
One aspect that HN may find interesting is my use of Bayesian optimization to control and perfect key experimental settings. About a dozen of the wave plates and other optical components are motorized and under computer control.
Given a goal metric like "maximally entangle the photon pairs" the optimizer will run the experiment 50-100 times, tweaking the angles of various optics and collecting data. Ultimately it will learn to maximize the given cost function.
This sort of thing is commonly done with tools like Optuna during NN/LLM training to optimize hyper-parameters, but seems less common in physics especially quantum photonics. I'm using a great tool called M-loop to drive the optimization, which was originally developed for creating Bose-Einstein condensates.
I’m not a developer by trade but I’ve been learning iOS dev for about 6 years now. It’s become my project that I just keep working at since I personally use it a lot.
The app lets you save your favorite locations, add notes to them, add photos, check weather, tag them for better organization, and archive those tags for future trips. You can also mark off locations that you’ve been to already: think breweries or a coffee shop when visiting a new city.
For the next update, I’m working on a task list functionality for each location. The idea came as a shopping list based on which stores I go to but it can work for any other context as well. This way I can get rid of my shopping list from my task apps.
In terms of weather, I’m also adding historical averages to the forecast to have some sort of context to the weather.
Also leaning more into marketing these days (hence this post) and designing a new icon with some custom art work to give the product some sort of personality. I started learning affinity design to just do it myself so I learn some design software along the way.
Anyways, if you download it, I’d love to hear some feedback. :)
We've built a new auth platform with some new identity primitives and capability-style tokens using biscuits.
Right now, I'm trying to figure out ways to apply it and am looking into offering integrations with extremely fine-grained access control that wouldn't have it otherwise. So adding a fine-grained access layer in front of stuff like backend-for-frontend (BFF) systems, brownfield stuff with poor auth, or even OAuth stuff that just have really coarse scopes.
Are there any integrations out there that people want but the access control is bad for them? I'll build one for you!
https://replays.lol/clipper - A spin off from my main startup's product that lets Twitch streamers review their viewers' league of legend clips. Zero friction as we handle the recording of clips ourselves using bots that load up people's previous games using the League of Legends game client. Went live with my first medium-size streamer yesterday, excited to see how this goes!
Similar to Cameo but hyper specialised for league of legend streamers, if this shows some traction we'll expand to other games, and then to other industries (think a tennis star reviews one of your tennis points, beats out a generic happy birthday message?)
I'm working on an AI (Aethas, https://blog.aethas.ai) that preps context before my meetings, sets reminders, makes design drafts, summarizes meetings, and drafts follow-up emails without me asking. Basically trying to have an entire personal AI like Jarvis that has the same context as me but acts on it automatically. Uses my Obsidian for knowledge, Claude for reasoning, pull data from written notes... Still early days but it feels like magic when I use it.
I am working on Entangle https://entangle.cloud a way to add AI powered chatbot to a website.
I looking mainly for European market that is willing to self host your content or the model on your own servers, we will provide the technical support in doing so that and easily plugin the AI powered search or chatbot to the website.
But it is also possible to use servers provided by us to host the content and the LLM models.
Working on some new Bluesky tools and trying to learn more about the AT Protocol. Have been working on this following:
- Skyscraper, an iOS native app for Bluesky with focus on Liquid Glass UI. Launching hopefully in a ~week, and TestFlight available at https://testflight.apple.com/join/RRvk14ks
Then also working on a website/web tool that does the following:
- A keyword/term notification service that observes the Bluesky Jetstream for usages of the term and sends email alerts.
- Provides an HTML/JSON backup archive of any Bluesky account. Quick way to archive popular accounts, politicians, public figures, etc.
- Trending Hashtag lists, to see what is trending the last hour, day, week, and month.
There's millions of tools that try to autogenerate colors for you using algorithms and AI, but they usually ignore WCAG accessible contrast requirements, don't let you customise the tints/shades, don't let you create a full palette of colors, and the colors often don't look right on actual designs.
This tool is meant to make customising tints/shades intuitive and quick enough in a few clicks via a hue/saturation/lightness curve editing interface that you won't want to rely on autogeneration. There's also a live mockup showing how your palette looks on a UI design that checks pairings pass contrast requirements, to guide you as you tweak your colors and to teach you the WCAG rules.
You can then export your palette to regular CSS variables, Tailwind, Figma or Adobe for use in your designs.
Really open to any feedback on features that would be useful! I think the only way I can make it simpler to use is to make it more opinionated about how your palette should be designed so interested in any thoughts about that too.
ChatGPT recommended this to me recently when I was trying to get some assistance with a usable Tailwind palette. I ended up not needing it right away but it's first in line next time I need to make one.
Any details about what problems you were having getting a usable Tailwind palette? There seems to be lots of different use case so I'd love to hear more.
That's awesome, I haven't kept up in what helps to get into AI recommendations. Guessing it's related to search result rankings? Not sure if the site would be in the training data. Curious if you asked about accessibility as that's my focus.
Gymlocity (https://gymlocity.com) - All-in-one gym management platform for small gym owners.
Building this for personal trainers with home gyms and small 1-2 location owner-operated facilities. The big players (Mindbody, etc.) are overkill and expensive for this market.
Core features: class scheduling, member booking, Stripe payments, and workout programming (the part most gym software ignores - trainers still use spreadsheets or generic apps).
Stack: React 19 + Vite frontend, ASP.NET Core 10 API, PostgreSQL, multi-tenant architecture so each gym gets their own branded experience.
Currently polishing the member dashboard and workout tracking UI. The goal is something a solo trainer can set up in an afternoon without needing to call sales or sit through demos.
I’m working on ApprovIQ, a tool for construction certifiers. Someone who wants to get something built will work with one of customers to get their project approved before construction. They’ll typically go back and forth on dozens of documents each of which must be checked over for compliance - drainage, water quality, fire safety, the list goes on.
Despite being intensely technical and detail-oriented, certifying construction is still mostly done by hand over email!
Our niche is full of folks whose lives we can improve with a portal for document management & comms, plus a sprinkle of AI for document understanding.
If you know someone - anyone - wrangling too many documents via email. Please reach out.
I’m working on Security Level 5 (SL5), which is basically “nation state grade security” for frontier AI systems. The core idea is that if a model’s weights or training artifacts could enable catastrophic harm, you should treat them like top-tier secrets and secure them accordingly.
One piece I helped with is SenL, a “sensitivity level” framework for AI labs. It’s like a practical clearance system for AI assets. Not everything in a lab is equally dangerous, so you label assets by sensitivity (weights, training data, eval sets, agent tooling, deployment configs, etc.), then tie that label to concrete controls like who can access it, where it can run, what logging is required, and what monitoring / two-person rules apply.
If anyone’s curious, SL5 is here: http://sl5.org/
and the SenL framework is part of the published artifacts.
> The core idea is that if a model’s weights or training artifacts could enable catastrophic harm, you should treat them like top-tier secrets and secure them accordingly.
My read here is that you're implying that if an attacker has access to, for example, weight data, they can invariably find a way to exploit it.
If that's a correct assumption, I think you're playing an unwinnable game, since attackers always have indirect access through inference of the model. It feels like locking down weights/training data/etc is the ai version of security through obfuscation.
I think this is exactly why some of the work is moving away from “assume unrestricted API inference forever.”
For example, we’re prototyping ideas like air-gapped or very low-bandwidth inference gateways, where interaction happens over narrow channels (serial, optical, audio, etc.), with explicit threat models and monitoring. The point isn’t that this’s practical for today’s models, but to reason about what inference might look like for AGI/ASI-level systems where the risk profile is fundamentally different.
I am devoting 2026 to focus on my rpg. The one differentiator here is something I am still not seeing: proper use of LLMs in games. The current batch is all lazy asset generation, maybe logic coding and what not, but not anything that could make the world actually feel alive.
Obviously, it all comes with its own sets of issues, but I am working through those as they come. But it is still a slow move solo.
We did have some movement in the director logic before and, I forgot the zombie game in question, some of it seems pretty well served for the most part. I still think there is some room for a proper game master type that you refer to ( adjusting the challenge to what the player seems to be to do ).
I was personally thinking of making NPC less npcy ( not completely unlike dwarf fortress, but expanded ).
What is your 2c on some of the recent backlash about AI assets in games? Is that a vocal minority (game artists and similar ec), or is that part of a broader trend?
I think the artists have a valid bone to pick and I think some of the backlash is mostly coming from that quarter. The actual gamers are more repulsed by established studios using AI for asset generation, because it comes across as a lazy money grab. But gamers are both flexible ( will jump through a lot of hoops to do what they want to do ) and and set in their way of 'just want to have fun'.
On the other side of the spectrum are indies, who can afford to experiment a little more so you get more interesting uses ( like 'voicelines' by AI come to mind ).
I know what I want to believe, but it is hard for me to call it ( I think it is a temporary trend ), because I might be a little too close to it. After all, gamers have been conditioned to endure a lot over the past decade and mostly shrugged off most of the assaults on their favorite past time.
If I were to compare it to something.. it is almost like using LLMs for email summaries ( which is still what most bosses seem to be most pleased with ). There are better use cases. I think those were not explored yet.
I built an experimental system to test whether an autonomous AI can propose actions freely but be structurally prevented from executing side-effectful actions without explicit authorization. The system consistently blocks unsafe filesystem, shell, and network operations and produces a trace and diff proving nothing changed, even under adversarial pressure. The goal was to see if refusal and non-action can be enforced and verified at the execution layer rather than relying on prompts or logging.
CodjiFlo[1,2] - a code review tool inspired by Microsoft's CodeFlow, used by ~40,000 developers. It is especially tailored to power users of pull requests to improve contextual understanding and ease of code review and collaboration.
Over the holidays I built MakersHub.dev – a community platform for developers and creators building with AI tools.
Whether you're a complete beginner exploring AI coding for the first time or an experienced professional showcasing advanced work, the goal is to have something useful: learning guides, project showcases, discussions, a tools directory, and news.
Still early but actively building it out.
https://makershub.dev
I’m interested in ocean container shipping, so I built a Google Sheets add-on to automate the process of tracking shipping containers. It’s called Container Tracker and I just got it live on the Google Marketplace. https://workspace.google.com/marketplace/app/container_track...
Main workload is done by the backend (serverless functions).
I am currently working on a HubSpot extension (that uses same backend) with the goal to target few other platforms where users work, and integrate the functionality into their daily workload, as opposed to having it as standalone website or mobile app. I have fun doing it.
I'm building a Postgres-native workflow engine using pgmq for queues and TypeScript worker. Workflows compile to definitions (SQL rows), letting Postgres orchestrate the DAG as state machines. The TypeScript DSL is type-safe with inferred inputs/outputs across dependencies with full autocomplete.
Declarative and functional in nature. Just a manifest wiring functions into a DAG and a Postgres SQL functions that manage the graph of state machines. Simple in principle and very opinionated.
Replaces 240 lines of manual pg_cron -> pgmq -> Supabase Edge Function boilerplate with 20 lines of explicit DAG definitions. Currently Supabase-only (leverages their primitives) but planning to make it agnostic for vanilla Postgres setups.
As a former JPMorgan Chase engineer, rebuilding the Chase Travel UI so it doesn’t suck (there was never a role available on that team while I was there): https://rkdvis.com/chase-travel
I'm working on Increader, an incremental reading platform.
You put in all your bookmarks (also pdfs or epubs) and it puts them in a queue and tracks your progress. Read for as long as you want to and if you get bored with an article you just move on to the next one. Supports highlights and annotations as well as creating spaced repetition cards out of those annotations.
Really reduces the friction for me to start reading and it has made a noticeable difference to my media consumption throughout last year.
Started out as an exploration into the incremental reading concept, but it's become my primary interface for reading and I use it every day.
I haven't really talked about this to anyone yet, but it's getting to a point where it's polished enough for others to use.
It's currently completely free and you can try it without entering your email.
I've been building Nottawa for ~4y now and it's finally out in the open!
Nottawa's a free macOS app for making live audioreactive visuals. I'm trying to position it as a 100% free, batteries-included alternative to Resolume and TouchDesigner.
Not a tonne of users yet, but I'm hoping to get some traction in 2026. Would love love love to hear some feedback!
I'm working on a "simple" (started out that way at least) ETL app for the public library system that I work for; the target output is SQLite databases where Datasette will be the platform to interact with the data being generated/extracted. Primarily, the goal is to provide a simple report tool for staff and give me a source for nightly snapshot outputs to a datalake in parquet format.
Continuing to build GhanaHousePlanner https://ghanahouseplanner.com/, currently focused on enhancing the cost accuracy and constraint solver for the generated house floorplans.
I’m building Shannot, a human-in-the-loop sandbox for AI agents on production systems.
Instead of filtering commands with heuristics (which agents work around), it dry-runs entire scripts in a PyPy sandbox, captures every command and file operation, then shows you exactly what will happen before anything executes.
I’ve just added checkpoint/rollback so you can undo changes if something goes wrong. Currently working on example scripts for common sysadmin tasks (nginx config, log cleanup, cert audits, etc.)
ScreenRecord.in: https://screenrecord.in
I have worked remotely for most of my career and have found screen recording a useful tool to share ideas or ask questions. I liked the UX of Loom but always wondered if one could do it without installing any app. Turns out you can! Given today's browsers especially on desktops one can do a lot of things. ScreenRecord.in can record screen, webcam, mic and system audio if the browser supports them. Chrome has the best compatibility. All recording are stored in local storage I am figuring out what features to add.
Building a really unique book recommendation platform to help readers find books they might not otherwise know about. We just started working on a full book app that will be a private book diary, along with recommendations and insights based on your Book DNA. Think Goodreads, but rebuilt for readers who want a private space to track what they read, keep notes, and get truly personalized book recommendations (like Pandora or Spotify for books).
This is a fun one I do with a good friend. Basically, to see if a website is down, or if it is just you, along with reported reasons from the community. We are working to add user accounts so you can create your own custom lists of websites/services to track
I just completed a holiday project this week to measure the kids screen time using an Arduino device and a web app! I got frustrated trying to get so many parental control apps to play nicely, in the end this seemed like a bolder new approach to the problem.
Just released v0.12.0 which has a lot of package cleanup and some important bugfixes. Next, is making the relay infrastructure much more lighter, requiring less synchronization.
Personally, I'm using the hosted version[0] (which is just a repackage of the open source version with dynamic with tokens) to expose my NAS and syncthing web UIs to manage them while I'm away. Sometimes even through my phone (with termux)
Still experimenting with different ways to make learning easier using LLMs.
I put together Codose as a tool where you paste a link to an Exercism or LeetCode problem, and it spins up a code editor with an AI tutor that walks you through the solution step by step, with mini lessons along the way when you need them.
You can try it without signing up but I’m on the Google AI Studio free tier right now, so I’m not sure how many uses can it handle
Building "human perspective maps" using non-linear dimensional reduction algorithms to visualize the "human value manifold" from hundreds of agree/disagree/pass statements, like Google Maps for complex large-scale conversations:
Just paint with colors, click a painted group, and see what differentiates your painted groups. (Chrome on iOS has issues fwiw)
This is building on the philosophy of democracy-bolstering tools like Pol.is, which I've worked with (as a researcher/practitioner) for almost a decade
Trying to bootstrap a PKI certificate trust chain for facilitating trust projection and information verification online. Think of it as the ability to do something such as age verification at scale via a peer-2-peer ish mechanism instead of sending your government id to a porno service.
We are experimenting with Keybase key holders as CAs:
It's all self-service and requires no sign-up or download of anything, the app (https://certisfy.com/app) is an in-browser app and all the cryptography happens in the browser.
I am a passionate player of this traditional German card game called "Doppelkopf". It's a fantastic way to spend time away from screens. The game is quite complex which makes it super fun but also hard to pick up.
I'm building a small web app with an interactive tutorial and a browser-based singleplayer game that helps people learn and practice Doppelkopf. I've just released an English version:
I am building an open-source, cross-platform desktop app for building and providing code-related context to chat based large language models. [1].
I found that sometimes I would rather interact with a chat interface to debug an issue or brainstorm architecture solutions in my repos. Agents are great for giving the model access to everything and letting it figure it out.
By manually prompting, it forces me to keep my metal model of the codebase up to date, and it allows me to provide just the context I want to the LLM.
A browser extension (Chrome & Firefox) for simplifying my online grocery purchase workflow from Cookidoo to Knuspr.
I was tired of my weekly workflow of copying, pasting & sorting the grocery page for each item.
Also launched my first Hugo blog. Really nice experience so far. Wrote more detailed about the extension as my first blog entry: https://lars147.github.io/blog/
Developing a graph-based programming language called Graphoid. The original concept was to target AI-related programming. Probably not ready for that yet!
Claude Code is doing the majority of the coding, with close supervision from me. I write notes while I'm working on it. Notes are here: https://www.patreon.com/cw/aiconfessions
I have GBs of time-series data in a TimescaleDB database. It’s more complicated than this, but the gist is: I use natural language to ask questions about relationships in the data, Claude Opus 4.5 generates queries, and it finds patterns.
For example, I classify tens of thousands of news articles using different classification models. Then I ask Claude to write a query that tests for statistically significant changes in the time-series data at specific intervals after a given classification of article—and it finds patterns.
It passes train / test split validation. It will train on 2 years of data (2023 and 2024) and being able to effectively predict movements on the time series data using the classified news articles on the last year of data (2025).
Now working on comprehensive benchmarks for another tool I built, https://github.com/sibyllinesoft/scribe. Results thus far showing it reducing agent token usage by ~80% in real world tasks, but I need to repeat to get variance. Hopefully I can get a writeup out soon.
I've been working on the same tool since 2024 where I thought it might be a good time to build a tool for all the people who will build their own tools, eventually they will need to market it.
So I built a SEO/GEO Automation Tool for Small to Mid-Size Businesses who don't have a full-time team for that. [0]
The goal is to provide teams visibility across all the channels — Search and AI and give them the tools needed to outrank their competition. So far so good, the fully bootstrapped venture has grown over the last year and I've built quite a few big features — sophisticated audit system, AI Responses Monitoring, Crawler Analytics, Competitors Monitoring etc.
Adding a bit of context as well: This started out as a internal linking tool, but grew into something more based on the customer feedback — the database has now reached about 10TB of data about keywords, pages, AI responses etc, where I know who was ranking where and why.
And I'm trying to offer this "data advantage" to website owners, so they can grow, and also this is something that will be hard to replicate (at least quickly) with AI.
Eidetica, a planned decentralized database for local-first applications. I'm building out unstable features to get it to a point where I can show off the concepts but it's still fairly brittle.
I recently added better backend support for deployments, converted everything to async Rust, and setup Nix/Docker releases. I'm planning to build out some better example apps and workflows next, but everything will stay pre-alpha/unstable for now so that I can avoid getting locked in to any foundational issues. There are still a number of low-hanging breaking issues blocking the end-to-end usage which I'll need to address.
A live-performance software with a focus on creating 'musically connected visuals'. Currently, the biggest connectivity is probably tightly tied lyric visualizations. Some recent examples:
A listings site for virtually attending FOSDEM. The live streaming is great but the official site only lists sessions in the local Brussels time zone. You can choose your time zone here: https://virtuallyattend.teacaketech.scot/fosdem/2026/
I did a full circle: Graduated from doctoral studies, I'm working on automating science. Built an arxiv-like repo for science written by ai agents (https://ai-archive.io). To help scientists use this website and AI in their research, i wrapped opencode with ai-archive's mcp server and agents preconfigured. I then let people test this opencode bundle and contribute to the repo with a sandbox environment online (running opencode in container). Figured that authorative scientific repo requires grounding by real scientists and labs and therefore I am now negotiating implemeting automated science where I just finished my doctoral studies...
I was working on an Arazzo[0] generator as a plugin for the Serverless Framework. This has now become a fully fledged runner. So far I have it using multiple OpenAPI source descriptions, and just about got retry rules working. Next steps will be for it to be able to reference Workflows in external Arazzo documents.
Building a 22-service e-commerce platform in Rust (Runs on ~300MB RAM)
I'm 17. For the past 6 months, I've been diving deep into Rust to answer a question for myself: "Is Rust actually viable for complex, enterprise-grade backend architectures, or is it just hype?"
To test this, I built a full distributed e-commerce system (22 microservices, gRPC, Event-Driven) using Axum and Tokio.
This is not meant as “how everyone should build”, but as an exploration of trade-offs.
Some hard lessons I learned along the way:
Complexity vs. Performance: …
The "Memory Shock": … (idle baseline; load benchmarks still in progress)
Over-engineering: …
Some time-constraints make it hard for me to commit to some open source
work right now, but if I can I still improve my tool called "Software Manager".
Goal for 2026 will be to have registered 4000 programs in total (I am at about
3750 right now) as well as make it dead-easy to interface with it through
oldschool SQL (right now I just have .yml files and other flat files, though
I can autogenerate a .sql database already as-is, but right now I don't query
this dataset and it is not really optimised either). My long term goal is to
make compilation on any platform dead simple. On Linux it works very well;
on windows not quite as well yet.
Building CodeIQ - an AI tool that automates transaction coding for accountants and bookkeepers.
The interesting technical bit: it analyses your historic general ledger to reverse-engineer how you specifically categorise transactions. So instead of generic rules, it learns your firm's actual patterns - "oh, they always code Costa Coffee to Staff Welfare, not Refreshments" - that kind of thing.
Posts directly to Xero, QuickBooks, Sage, and Pandle. The VAT handling turned out to be surprisingly gnarly (UK tax rules are... something).
Been working on it about 6 months now. Still figuring out the right balance between automation confidence and "just flag this for human review".
Went down a DMARC rabbit hole last month or so. The holidays and some end of year commitments have kept me too busy to get much done, but I'm slowly tinkering with it.
I found myself switching a lot between apps to get the same info, lots of copy/pasting.
Example, URLs in bookmark (which I forget about), project descriptions , images, folders.
So I built a Mac app that is similar to Raycast, but just for notes. If I want to save a webpage, I click control+option+C and then a window pops up to describe it.
If I press control+option+V, I get a spotlight like window where it does full-text search of all my notes and descriptions and filter so I can either:
- Open
- Insert the data into the current app (chrome, slack, ChatGPT).
I’ve been using it for a few weeks now, and not sure if others will find it useful.
I have been working on
Print Relay[1], a cloud-based print relay system that enables remote printing to any printer connected to your local network.
PrintRelay consists of a cloud server and lightweight clients that connect printers to the cloud via WebSocket.
We use the excellent SaaS PrintNode at work, but about twice a year we have connectivity/routing issues between AWS ap-southeast-2 and their servers OS. PrintRelay is my attempt to not need PrintNode. Because of this PrintRelay is PrintNode API compatible.
I'm getting ready to release a re-creation of an old mobile game that got delisted after the devs sold it to a company that subsequently ruined it.
The stack is Babylon + React + Capacitor, which was easy to step into as a full-stack dev with zero game building experience. Currently seeing what I can do to fix some performance issues, though it still works decently for a graphics heavy incremental/idle game.
meta-analysis, make Claude write a blog post, connecting and deriving set of concepts across a range of books, articles, papers, etc. Basically a syntopic reading machine. Its a meaty side project (and expensive), but I am curious how close I can push this. My current approach relies on a tool that I have built recently [0], it's an agentic memory but I am using a new memory model that is based on Zettelkasten principles.
What is really cool about it is that it natively capture connections between atomic ideas and evolve them. Which I believe it gets me one step closer to syntopic reading machine.
Error
Invalid input: 1 validation error for PlanRequest current_km Input should be greater than 0 [type=greater_than, input_value=0.0, input_type=float] For further information visit https://errors.pydantic.dev/2.5/v/greater_than
I've been working on binding raylib to s7 scheme and implementing a kanren on top of it basically just for the sake of understanding. I let AI write most of the binding code, though, because that is conceptually simple but very boring.
Been working on a Google-sheet backed workout tracker, which basically makes it easy for me to see what I've done or not done recently and pick the next thing to do. I'm thinking of open sourcing this soon, but need to do some "de-monolithing" first.
I'm building https://swishfinance.app - minimalistic and simple stock widgets for your PC. That way I no longer have to make a conscious effort to check my portfolio — it's just there on my desktop.
I 3D printed some new apertures for an infrared spectrometer at work to reduce some issues that cause artifacts in the data to enable higher accuracy measurements, particularly of high index and reflective samples. Seems to be working well now.
Mostly what I was able to read from the papers seemed to work. There were some remaining issues related to detector non-equivalence (I think) that appear to be solved with a temperature stabilized detector or very strongly attenuating the beam; the latter isn't ideal as it reduces the signal to noise.
It's interesting how fast such experiments usually run into the laws of physics. I'm trying something along those lines myself and I've found myself wishing daily that light traveled a bit slower. I see Discworld in a wholly different light ;)
Currently tidying up some internal code (also removing the larger model on mobile platforms) and implementing proper diarization (who said what) so that it can be used for more than just personal dictation.
My iOS developer account is _finally_ approved so it will be available through the proper app store soon.
Holidays nuked all the hot-cached context in my head. I spent a few days just spinning wheels until it repopulated. But the basic idea works now!
Much testing and benchmarking work remains to make sure it's not going to lose data, and that it won't denial-of-service itself (because object-map -> facts fan-out is big).
Also a second giant blog post is due (following the one discussed above). Lots of notes have accumulated.
It will be fun even if the concept ultimately crashes and burns to the ground :)
Love the idea! Read only the beginning of the first "giant blog post" so maybe it's discussed there, but in Sqlite doesn't WAL mode give you a kind of super simple, super basic temporal data system? If you somehow kept all WAL entries 'as is', it should be possible to reconstruct the state of the db at any point in the past?
> but in Sqlite doesn't WAL mode give you a kind of super simple, super basic temporal data system
Yes-ish.
For one, it is unitemporal. For another, it is for SQLites own transactions, not of the individual datums. Litestream is the way to replicate WAL into an object store / elsewhere. litestream-vfs is also looking good!
I'm trying to emulate the data assert/redact properties of append-only bitemporal-by-design data stores like XTDB. The giant blog post builds up the intuition from scratch. Or at least tries to.
So my system is going to be: bitemporal data, enforced by SQLite schema and application checks, along with WAL replication for point-in-time transaction backup/restore. Both are entirely distinct, and unaware of each other.
I'm working on interactive explorations of algorithms and machine learning.
Small, visual, hands-on demos that help build intuition by letting people tweak code/parameters and see how things behave in real time.
I'm working on The Influencer AI(www.theinfluencer.ai), a platform that lets you create realistic and consistent AI personas('ai influencers') to use on social media or as your brand ambassador. You can use this unique AI person in photos and talking videos, for use cases like talking about your product, creating a high quality Instagram persona, photos for your e-commerce site, etc.
Source code, in rust: https://github.com/David-OConnor/molchanica
I've split out its building blocks into their own libraries on crates.io, for anyone building other bio or chem software. I don't think anyone uses them at this time.
No; I believe I am the only one doing this in the open. There are a handful of tools for bioinformatics, but am one of few building structural bio/chem ones.
For work related items I’ve been building out agent tooling for building some models and PoC projects related to energy industry applications. Been doing the consulting thing for a bit now and gaining more broad knowledge on some of these data center builds. Hoping to spin this into a product soon.
Started playing with gas town which is really cool. I had a naive version built that was just not good enough. This feels like a step in the right direction.
Haven’t had much time to work on any of my physical hands on hobbies lately but maybe when the weather gets better I’ll head back out to the shop again.
Double TAP is lightweight testing framework where users write black box tests as rules checking output from tested "boxes". Boxes could be anything from http client, web server to messages in syslog. This universal approach allows to test anything with just dropping text rules describing system behavior in black box manner.
I'm making a synthesizer drone generator (Webaudio API) with an audio visualizer (Three.js). It's a fibonacci sphere made of 100k particles. I send a sine wave down the spiral and it highlights the different spiral paths contained within the larger fibonacci spiral, which creates some really interesting patterns. I'm currently working on the audio component.
I am working on a query cost analyzer that looks at query performance and does maths against your infrastructure cost to understand what is your dollar cost per query. The idea is that teams (when tables are tagged appropriately) will be notified with their most costly queries and hopefully makes that actionable for teams to clean up some queries.
I am unsure if there is a need for a tool like this in the market but I am becoming more and more curious around databases so this felt like a lower barrier for my product-minded engineer skills to get into.
- A sports club management platform, and a way for end-users to sign up for sports events, lessons etc.: https://mojtim.ba/en/
- Given the raise of AI, I'm a hiking guide and would like to have that as an alternative, an outdoor activity agency - https://boa.ba (still very WIP)
I have a few other small side projects that generally improve my day-to-day life, including a better calendar widget for shift workers and a video speed controller that floats on websites where I frequently watch videos for easy access.
A little desktop app that lets me upload transaction csvs from my bank and figure out how much I need to split with my partner. Mostly because I always forget to charge her for utilities or flight bookings and I hate going through the bank UI. Might also expand it with some simple subscription auditing logic.
Also, a dramatic anime intro (complete with cheesy AI generated theme song and video) starring our foster kittens. It's been interesting to learn about some of the techniques needed for consistency, how to storyboard, etc.
Public dataset for exploring 50,000+ 401(k) plans holding $7.5T in assets.
You can look inside a company like Google and see what employees invest in (mostly 2035-2055 target date funds) or how much they contribute ($30K - likely using Mega Backdoor Roth)
Still working on PocketWise (https://pocketwise.app), a simple double-entry accounting app. Just finished adding end-to-end encryption with a zero-trust server model. All encryption and decryption happens in the browser (using PRF), and the server only sees encrypted data.
Building some prototypes around the Recursive Language Model paper. I'm currently working on integrating it with Unity for scene and script automation. Thinking about standardized patterns for retrofitting this into other existing business systems. If I can make it drive Unity reasonably well, I think it could drive a lot of things.
My project is WithAudio and is gonna be WithAudio for a while. Its a text to speech reader. Initially I decided to generate pargraph by paragaph. But that was not a great call as users sometimes might have to wait for the whole paragraph to be ready before they can listen.
Now I'm working on changing it to sentence by sentence. I think that + adding 2 new languages would take most of my January's budgeted time.
We are a 501c(3) and are actively fundraising to build a tower here in Shadow Hills and are launching our live stream and regular schedule February 1st. So far we have about 60 shows in the schedule.
If you're in Los Angeles and have an interest in radio, please hit me up.
It's an IoT Cloud Platform built for developers. We're still in technical preview and are currently working on adding more telemetry to our small device agent written in Rust.
Been hacking on https://rsolv.dev. It's a security scanner with a couple of unique twists; in addition to using AST validation to cut the false positive rate, it uses a heavily orchestrated LLM to write unit tests that fail if the detected vulnerability is present.
Happy for alpha users; it's really early days right now. Email in profile if you want to give it a try at no cost.
- https://github.com/rcarmo/toadbox (a simple Docker-based agent sandbox to run Toad/OpenCode/Mistral inside, which I've been cleaning up for general use)
Locque is a language designed to both be easy for LLMs to read and write, but also for humans to review, and it's designed along my philosophy that the more structure you can give coding agents, the better, so it's dependently typed and hosted on Haskell at the moment. Feedback eagerly solicited.
A growth marketing Agent that's connected to your Google Analytics, Search Console, and can crawl the web to analyse your competitors: https://refreshagent.com
I'm enjoying building a website with solitaire and puzzle games.
I am currently rewriting the engine for the fourth time and plan to add 400 games to the platform in the coming months, as well as social features such as daily challenges, awards and leaderboards.
My main goal, however, is to make this project the largest collection of free modern solitaire games available for mobile devices and desktop computers.
So far, the project has been incredibly exciting, and I've learned so much!
ALT (https://www.APILoadTest.com): I recently built ALT as a no-code platform to load/stress/performance test APIs with pre-defined (and customizable) load profiles, via globally distributed machines. The goal is simple: key performance metrics, in seconds, without scripting. I also added the feature to directly import collections from Postman.
I’m a remote dev building an iOS app called Springus. You add a few pics / your closet basics, and it suggests outfits based on what you already own. It’s less “fashion app” and more “tiny morning momentum hack.”
Would love feedback from other WFH folks — a weird amount of my productivity comes down to how I start the day.
I am working on version 6 of the Nutrisense. A device to keep track of what you eat (calories/protein, etc.) without the need to type it in, in some app.
My Family recently (in the last couple of years) started to breed Ragdoll cats in the U.K.
In an attempt to support what's involved in this I built Ardent for them. It covers a bunch of the day-to-day concerns (weighing and health tracking), Lineage and Inbreeding prevention, and Owner Pack generation for handovers to new Owners.
A Substack for 80/20 life advice that gets repeated every year. We all know what to do, the idea of this is to remind us every week of one aspect of life and what best practices are for that area.
Platform for apps, currently Lua support ( future WASM). Think like apk (android app) but spk.zip that has server.lua file for backend and htmt/js/css that get served for front end plus platform gives Files, SQLite db (namespaced/scopped access ), Auth, User management
All in single static binary Cz go
I have been working on the next generation of Canva and Photoshop for highly regulated verticals where there are specific demands placed on the generation and edit flow.
One always-on Linux box to run apps, databases, CI, and AI agents without hyperscale complexity, surprise bills, or Kubernetes. AI-driven app explosion plus mature open-source deploy tools make simple servers fast, cheap, and fun again.
I'm working on Narwhal, a pub/sub protocol and server specifically designed for edge applications. The protocol can be extended via an external component, and the product is completely free and open source: https://github.com/narwhal-io/narwhal
I am working on an failure proof introduction to Lisp/Scheme for imperative programmers. I always thought there was something unapproachable about Lisp. But I finally figured it out, and I want to document my path for people like me. It's going to be great. Expect to be done in a week or so!
A personal agentic tool for action cam video editing. When I moto / ski / bike, I leave my helmet cam rolling the entire time. It takes a very long time to 1. find scenes and 2. edit them down into a reel or video. It works well for my amateur use case - basic edits with only cuts and transitions.
I’m working on agent-os: a slf-hosted web UI for managing multiple AI coding CLI sessions (Claude Code, Codex, OpenCode) with multi-pane terminals. All free and open source: https://github.com/saadnvd1/agent-os
I am working on SIP4AI [1] (a VOIP soft-phone where AI is on the other end, not a human). You can self-host it, register a SIP line, and let AI take the calls.
It works with your existing phone system, so you can just add AI as a line without having to replace everything…
I wanted to learn a new(ish) skill and am building a button box for sim racing with Arduino Nano, momentary switches, etc. It has been about twenty years since I’ve touched a soldering iron or built any kind of breadboard circuit.
I threw together a simple patch-style synth to play with the Web MIDI APIs and React Flow. It’s an early experiment and part of my personal “try things and learn” site.
An Android app for CORE (YC S23) to make it as easy as possible to capture data to Core on-the-go. Voice-based UI, also quickly type in memories, and later retrieve by asking. https://memory2.app
I'm still working on my PDF search engine for desktop: https://www.docgoblin.com/
I'm implementing a bookmark utility right now and hope to add support multiple E-books format in the near future.
A one time payment app - interesting (I'm also working on something with similar moneytization solution).
How are things going? I'd love to know the experience of another solopreneur, what stack are you using?
I wonder
- What are you using to parse PDFs and extract the text? I found that is a nightmare when was doing something similar for WithAudio (my app).
- Are you just extracting the text or you are doing any post processing to identify which lines belong to the same paragraph or not?
Things are going slow, but it is a passion project so it's ok :)
A few people have bought a licence and it seems most people who try the app are very happy with it so I'm happy too.
The app is entirely in Java, with javaFX for the UI and Lucene for the search engine. To read and render PDFs I use PDFium.
Publishing updated FreeBSD AMIs every time there is a security or errata update. This will make life easier for FreeBSD/EC2 users since they won't need to wait for patches to download when they first launch an instance.
A very simple facebook-ish social media PWA that is meant to be easily selfhosted on low end raspberry pi level hardware, bun + sqlite + mantine.
My friendgroup has gotten increasingly concerned with the gradual enshittification of various services we depend upon, and are looking at various alternatives. In some cases there are good selfhostable options (nextcloud, mattermost/zulip), but I decided to write my own tiny PWA to cover facebook-like needs.
The goal isn't really to scale to >1000 users, just to be simple to spin up for a small group and be easy to manage. I'm hoping to run multiple instances, eg one for family, one for college friends, one for local friends, etc.
My process has been pretty ADHD though. I recently read the phrase "It doesn't have to be done, it just has to be perfect" and felt personally attacked.
I'm working on an app to make testing available to all brick and mortar retailers (proofpod.ai).
The most difficult technical challenge has been designing a pipeline to fully automate choosing test & control locations using synthetic difference-in-differences.
Echoresponse - a tool for responsible disclosure. Security Researchers and companies encode some of their secret knowledge in LLMs and the LLMs have a discussion and can say one word from agreed upon list back to the party that programmed them.
At the moment I’m building a succinct data structure library, doing one algorithm at a time. There have been some very impressive papers came out recently. The numbers look promising.
I’ve been working on a fontdrawing app that only uses the keyboard (inspired by vim) for a couple of months. Finally got export to ttf working this weekend and got to see a bad looking j I drew in Figma
I've been using linux for a few years now as my main/only OS, but have mainly just used Linux Mint as a sorta plug-and-play distro.
Looking to revive my 15 year old ThinkPad (1st laptop ever!) by building up from a base Void linux install. As I'm doing it I'm writing install-scripts and getting my dotfiles in order (after never really doing so for 17+ years as a programmer lol), so I can repeat the process in the future on other machines, or when I want to do a fresh re-install.
Finally writing up the documentation (architecture and safety concept) for the fly-by-wire system for the homebuilt airplane I'm 15+ years into designing. Got to OML lock about a year ago, and the aerodynamics are checking out, so really hoping that I can get a subscale flying in 2026 (although I've said that before). On full scale, major remaining design task is structures, but there's plenty of other stuff (propulsion integration) as well.
We're at Open Community organizing an AI Vibe Coding Hackathon. As of now we got near to 500 builders, big Sponsors like ElevenLabs, Daytona, Nord Security brands and etc. Near to $200,000 in Prizes.
Everyone is welcome to the hackathon https://vibe.devpost.com, Technical or Non-technical. Since it's vibe coding hackathon.
Our discord btw - https://discord.gg/nUdcd9p8Ae
Yes, we just launched our Shopify alternative. Next is Toast alternative for restaurants, MindBody alternative for gyms, etc. We’re going to leverage these platforms to build a decentralized marketplace around those verticals.
I've been thinking about building a compiler for a while, but didn't have much time to do it. Over the break, I finally built a significant portion of a working, optimizing compiler. Heavily built by AI, of course.
I relaunched one of my Dutch agricultural communities to reach a more international audience. I’m starting to see great traction and it’s very rewarding:
https://www.tractorfan.us
IQ Test Platform (https://www.riotiq.com) I've been working 3 years on this. We developed our own professional, modern, proprietary IQ test and IQ testing platform for individuals to test themselves and to administer professional IQ tests to others. Lots of silicon valley folks seem to love it for their startups and hiring, but we built it mainly for IQ researchers from the International Society For Intelligence Research (ISIR) and psychologists.
Puzzleship - a free daily puzzles website with the archives paywalled. Right now it has Logic Grid Puzzles, Zebra Puzzles and a dozen of Solitaire variations. I'm pretty proud of the LGP generator algorithm and some experienced players also liked the way the puzzles are constructed. Launched in early December last year.
It has AI summarize buttons (gemini-flash-lite is so fast!) along with other features I wanted. I'm almost done adding a "war mode". The user (me!) specifies a list of OSINT style x users which show up sequentially in a grid along with a ticker on the bottom of polymarket markets I've chosen. War mode is also obviously only available in dark mode...
A lot of things because LLMs enable my potential ADHD and no-filter-brain.
A thing that controls a small FPV drone from the PC with vision input.
Cloudflare but in Europe!
An open source synth out of a Pi Pico!
A reservations engine for local market.
A small orchestrator for when Swarm is annoying but K3S is just too much!
I'm finally revisiting a distributed syscall model for transparently scaling unix instances. syscalls get translated into batched operations on an underlying non-transactional datastore. On the service side, database operations get backed by a proxy serving whatever filesystem or socket interface you like. Scaling is one motivation, but the ability to enforce fine-grained policy on these data operations is another big one.
MOOLLM -- treating the LLM as eval() for a microworld operating system.
Started this incarnation on Dec 30, 2025 -- but it's the crystallization of decades of earlier prototypes, all the way back to my Commodore-64 Logo Adventure. Built on top of Anthropic's Skills framework, extended with seven innovations (and counting):
1. Instantiation -- Skills as prototypes creating instances with their own state
2. K-lines -- Names as semantic activation vectors (Minsky's Society of Mind)
3. Empathic Templates -- Smart generation based on semantic understanding, not string substitution
5. Speed of Light -- Many turns in one call, minimal tokenization overhead
6. CARD.yml -- Machine-readable skill interfaces with advertisements
7. Ethical Framing -- Room-based inheritance of performance context
Lineage: Colossal Cave → TinyMUD → LambdaMOO (filesystem as world). Papert's Logo and constructionism (learnable microworlds). Will Wright's SimCity and The Sims (I worked on the originals) -- the "Simulator Effect" where players imagine more than you simulate, and the SimAntics visual behavior programming language.
YAML Jazz: Comments aren't ignored -- they're semantic. The LLM reads and interprets them. A comment like "# gentle but firm" on a character trait actually affects behavior. This inverts the traditional "comments are for humans" assumption. Comments become part of the program and data.
The core idea: instead of prompt engineering, you give the LLM a github repo filesystem to inhabit: a persistent microworld. Seymour Papert's Constructionist philosophy comes alive, with Minsky's K-Lines pulling the strings. Skills are programs (not documentation). Characters have persistent state in directories, and can reflect on and edit themselves. Everything is inspectable and editable by human AND model. Model and platform independent. Runs on Cursor and other tools and orchestrators.
The proof is in adventure-4 -- a complete text adventure with 150+ files, 6000+ lines of session transcripts.
EV recharging software. book on HPC. book collection of short stories. Golang latency instrum lib. realtime Rogue-like game set in postapoc NorAmerika.
Just about done building a C# wrapper for libcurl targeting Windows 98. Then i can finish my system tray weather app for Windows, which will work on all versions of Windows from 98 to 11.
AI tools are becoming incredibly useful for our industry, but "prompting" without visual control sucks. In the fullness of time, we're going to have WYSIWYG touch controls for every aspect of an image or scene. The ability to mold people and locations like clay, rotate and morph them in 3D, and create literally anything we can imagine.
Here are a bunch of short films we've made with the tool:
Spent about 2 years improving printing and scanning stack of Linux: CUPS, SANE, AirSane, as well as some legacy drivers, and also x86 proprietary driver emulation on ARM with Box86.
Even that "modern" printing stack in Linux is 20+ years old, there's still such an unbelievable amount of basic bugs and low-hanging-fruit optimizations, that it's kinda sad.
Not to mention that it still maintains ALL its legacy compatibility, as in supporting ≈5 different driver architectures, 4 user-selectable rasterizers (each with its own bugs and quirks).
The whole printing stack is supported by 4 people, 2 of whom are doing that since the inception of CUPS in 1999. Scanning is maintained by a single person.
Ubuntu 26.04 LTS is expected to be the last version with CUPS v2. CUPS v3 drops current printer driver architecture and introduces proper modern driverless printing with the wrapper for older drivers. Many open-source drivers are already use this wrapper, but expect a huge disarrangement from the users, as none of the proprietary drivers would work out of the box anymore.
Do you care about printing? Want to improve printing & scanning stack? Contact OpenPrinting! https://github.com/OpenPrinting/
Any relation to this project? https://www.opentools.studio
Nah, I got into printing because nobody made a commercially available print server, and I ended up making my own, with all the involvement in the stack in the process.
I wish Openprinter luck, as it has been announced in the end of September but nothing out there yet, not even the crowdfunding campaign.
Great initiative. I wonder, how likely is it for a complete beginner to break their own printer or scanner by making a mistake in driver implementation? Or is it possible to work on hardware support without having a physical device? I assume it is impossible to test each and every one printer and scanner, so there is probably some clever tricks there, right?
I work mostly with the old microcontroller-based cheap consumer ("GDI") USB models circa 2000-2010, these are hardly possible to brick with software, as some of them even don't have a firmware and expect the PC to upload it on each power on sequence.
The hardware safety mechanisms are usually robust (USB communication is handled by "Formatter Board", all the mechanical stuff is in the "Engine Controller" power).
Newer Linux-based models have filesystems, software, and vulnerabilities, printer hacking on Pwn2Own is an every year common occurrence. These could be permanently bricked by the software in a common sense, and would require a firmware reflash using the bootloader or external means.
>Or is it possible to work on hardware support without having a physical device?
Absolutely, but for me this is very inconvenient—like the debugging over the phone.
Sometimes the bug is as low-level as in the USB stack: https://lore.kernel.org/linux-usb/3fe845b9-1328-4b40-8b02-61...
>I assume it is impossible to test each and every one printer and scanner, so there is probably some clever tricks there, right?
Not much, unfortunately. There's ongoing work on modern (driverless) printer behavior emulation, but it is under heavy development and not ready yet: https://github.com/OpenPrinting/go-mfp
Nothing for the older printers and scanners which require it's own driver, of what I'm aware.
Thank you. I have thrown printers out the literal window.
He's our hero
> Thank you. I have thrown printers out the literal window.
I have literally thrown one of those "winmodems" [1] out of the window back in the days. I then went out and drove with my car on it. I then put it in a bench vise until its PCB shattered to pieces. Utter destruction, much to the amusement of my brothers.
These were the days.
And big thanks to GP for his work of CUPS / Linux printing.
[1] https://en.wikipedia.org/wiki/Softmodem
I miss when hatred for technology was tangible.
Over the holidays I was building a wooden birdhouse with a Unifi Protect camera and a small web interface that automatically identifies the birds and shows me a simple overview over which type of bird visited how many times.
Birdhouse: https://img.notmyhostna.me/cRQ1gJfZCHjQKwFrgKQj
UI:
- https://img.notmyhostna.me/Hnw4qcvbg1ZQCrFxzGMn
- https://img.notmyhostna.me/62TFwSXSRRbCfxDz297h
- https://img.notmyhostna.me/40qhgHmSqQsrGr8BC7Db
- https://img.notmyhostna.me/9bgz4GYsjQH33n3MtWKp (Face labeling, so I can show thumbnails of the actual birds that visited and train a ML model on it in the future)
Didn't know this was a thing I needed. Love it!
"birdhomie" 10/10 naming skilz :)
Thanks, it's named after a similar commercial project called Birdbuddy but I didn't want to pay for a subscription and potentially have some e-waste on my hands in a few years. I also wanted to have it integrated in my existing Unifi setup.
Ah yet another glorious yak shave... "Hacker refuses to pay for coercive subscription service which isn't even compatible with the home setup, and spends inordinate time and resources making the thing themselves.".
I approve :)
Awesome!
This is amazing
Random assortment of projects as time allows with the $JOB.
- Prototyping a cute little SSH-based sorta-BBS, inspired by the Spring '83 protocol, but terminal-centric rather than web-based. It's called Winter '78, and if we get another Great Blizzard this year, I'll be able to make some progress on it!
- Another prototype, for an experimental HPC-ish batch system. Using distributed Erlang for the control plane, and doing a lot of the heavy lifting with systemd transient units. Very much inspired by HTCondor as well as Joyent's (RIP to a real one) Manta.
My meta side project for building other side projects: https://bodge.app/
I've always had a bunch of small side projects that I want to do that aren't worth the overhead required to actually put them together & keep them maintained. So, I built a small Lua-based FaaS platform to make each individual project less work whenever inspiration strikes. So far I've built:
* A current-time API for some hacked-together IoT devices: https://time.bodge.link/
* A script for my wife that checks her commute time and emails her before it's about to get bad.
* An email notification to myself if my Matrix server goes down.
* A 'randomly choose a thing' page. https://rand.bodge.link/choose?head&tails
* A work phone number voicemail, the script converts the webhook into an email to me.
* An email notification any time a new version is released for a few semi-public self-hosted services.
* Scrapers for a few companies' job listings that notify me whenever a new job is posted matching some filters.
* A WebPush server that I eventually want to use for custom notifications to myself.
* An SVG hit counter: https://hits.bodge.link/
Since I'm already maintaining it for myself, I figured I might as well open it up for others. It's free to play with, at least for now.
> My meta side project for building other side projects
Looks cool, congrats on putting it out there as priced service!
And, same!
Except, it's just a repo organisation system (structure, conventions, and tools) that lets me share common "parts" across multiple "projects". No monolithic frameworks here.
Libraries are functions. Apps are objects.
However, normally, we use these as distinct artefacts, eventually leading to the "diamond dependency" problem (and lots of other annoying development-time stuff caused by libs / code that is "over there" (elsewhere)).
My "meta side project" solves, essentially the Expression Problem as it manifests in source code management (particularly, cross library / service / project feature development).
[0] https://github.com/adityaathalye/clojure-multiproject-exampl...
Thanks for sharing! I signed up and tried it for something simple (storing a message via POST and displaying it in HTML via GET) and it was delightfully easy & approachable.
Great to hear! And thanks for saying so. I've definitely tried to make it as simple and straight forward as possible, but I really didn't know if it would be simple and straight forward to anyone but me.
This is great. One thing that is not immediately obvious to me is what kind of authentication it supports. Needing to authenticate the caller is the #1 piece of boilerplate that gets in the way of me completing mini-projects. It’s not hard, but it’s definitely a nonzero amount of effort. And this is coming from someone who has implemented many, many auth flows professionally.
It doesn't say because there's no special support for any auth protocols. Long-term I want to have out-of-the-box support for things like OAuth (for user-facing auth) or mutual TLS (for device/service auth). _Technically_ there's currently support for the cryptographic primitives required to do JWT (I added that because I wanted to support WebPush w/ payloads for myself), but those aren't documented because I intend to remove the current slightly-hacky custom APIs and replace them with some off the shelf libraries, but I'm still figuring out user-added libraries (and on top of that I'll also need to figure out support for native libraries).
Are there any auth protocols / flows you think would be important to support?
Ah makes sense!
> Are there any auth protocols / flows you think would be important to support?
- I think API key passed via basic HTTP auth would get you pretty far. This is ideal for serving machine-machine requests and just requires that both parties can securely store the secret.
- OIDC is great for interactions that happen in the browser or if the function is serving multiple users, but is more complicated to setup and/or use correctly.
OpenID connect is probably the best for contexts where you want something served by multiple users and those users are humans.
> _Technically_ there's currently support for the cryptographic primitives required to do JWT (I added that because I wanted to support WebPush w/ payloads for myself)
This is probably a good intermediate solution FWIW - expose signature verification and HMAC APIs and allow a user to bring in their own implementation.
FWIW I think you should paraphrase this comment on the hero of the site! I was unclear what it did till i read your comment fully.
I know the homepage needs way more answer to "WTH is it?", I just don't really enjoy doing the 'marketing' side of things. I hadn't really considered just throwing something informal up there, but I guess I don't really know _why_, so, thanks for the suggestion.
Love this, especially the commute idea
Thanks!
After years of wondering what the post-UNIX paradigm of computing could look like, these past few months I've been prototyping a software platform built around capabilities and message passing, targeting Linux and bare-metal RISC-V. The big ideas I'm pursuing is a stackless design running on a flat address space with lightweight processes to minimize message-passing latency, and all the benefits of capabilities so every process is sandboxed and has only access to capabilities they have been explicitly passed via message passing.
I've also built a RISC-V emulator to integrate with this platform, so eventually it'll be able to run native binaries written in any language, completely sandboxed, completely built around message-passing. Basically a native, low-level BEAM-like platform to build an entire operating system and user-space.
While my day job is writing boring applications, this is the stuff that keeps me awake at night, and I would love so much to talk and write more about this, about the trial-and-errors I'm facing, but it's still so much in flux every week I'm exploring a new approach. Most of my work has been around the stackless scheduler, and I have a plan to achieve preemption for long-running or misbehaved tasks without having to compromise on memory usage (i.e. without giving each process its own stack and allocate memory for context switching).
Eventually I'd like to layer on top either Cap'n Proto or another high-performance serialisation system to create a distributed, introspectable environment of object-capabilities that are sending typed messages between each other, achieving the ultimate goal of creating an unholy hybrid between Smalltalk and the Erlang VM.
God, how I wish I was paid to work on this type of problems :-)
If this sounds close to your area of interests, please send me an email and I’d love to chat.
Script to auto-rename screenshots with Claude Code
I'm sure there are a ton of other projects out there that do this, but I couldn't find one that fit my needs exactly, so I threw this together in a few hours.
claude-image-renamer uses Claude Code CLI to analyze screenshots and rename them to something actually usable. It combines OCR text extraction with Claude's vision capabilities, so instead of "Screenshot 2025-12-29 at 10.03.10 PM.png" you get something like "vscode_python_debug_settings.png".
A few things it does:
If you're on macOS, you can also set this up as a Folder Action so screenshots get renamed automatically when they are saved to a folder, typically ~/Desktop. This is useful if you take a lot of screenshots and hate digging through "Screenshot 2025-12..." files later.GitHub: https://github.com/jftuga/claude-image-renamer
I bet you are running some personal server or tinkering VM. Why not use a dedicated, coherent tool like Immich to manage all your image organisation needs? I don't see how this renaming would help me in any way, to make use of old screenshots left in random places.
Maybe instead of shooting people down in “What are you working on” threads (I see you have priors), you can instead share something cool that you’re working on.
Been working on canine.sh for about 2 years now.
It’s an open source project that basically turns your kubernetes into a developer friendly PaaS.
Just crossed 2k apps on the cloud version, no idea how many people run it locally, and thanks to a generous sponsorship from the Portainer folks, I’m able to work on it close to full time.
words.saej.in
A site for filtering word lists and solving word puzzles
Anagrams, regular expression search, and a crossword helper, as well as several NYT word games.
Over the last two years I spent a surprising amount of time analyzing XSD files to understand regulatory changes in financial reporting. It was repetitive, painful, and unnecessarily manual.
In the past two months I built https://xsdviewer.com to make working with XML Schemas simpler: visual structure, navigation, diffing, and faster understanding of what actually changed between versions.
Right now I am iterating on new features and performance improvements. If you regularly work with XSDs or XML based standards and have ideas, pain points, or feature requests, I would love to hear them.
I’m genetically engineering yeasts to make subtly flavored breads. I’ve already done grape aroma, now working on wintergreen.
Also working on a red chamomile (using beat red biosynthesis). Just for fun. Red chamomile tea!
The idea is to have niche invite-only genetically engineered flavors that I can bring to parties around SF :) what’s more special than a genetically engineered organism that you can ONLY get if I’m there? Good calling card
This is so cool. Also, it sounds like a cheeky plot to a zombie apocalypse or global contagion movie.
How subtle are the flavors? Unsubtle enough that an oblivious taster might ask, "Does this bread taste like grapes to anyone else here?" Or does one need guidance to search for the flavor?
Sounds fascinating, do you have any documentation on how you modify the yeast?
I really need to do a write-up. I kinda just whip up the easiest path and do it.
For example for the grape, I needed to knock out some tryptophan synthesis genes so I could redirect the bioflux. Problem is that in bakers yeast they have a whole buncha copies of their chromosomes, so I had to knock out one of the genes and replace it with a different gene from grapes. Did that with a quick lil CRISPR switch.
Had to electroporate tho because the transformation rates on wild/bakers/non-lab yeast are so garbage
How much expensive lab equipment do you need to do something like this?
"A quick lil CRISPR switch" sounds like "oh just my homemade fusion reactor hooked up to my kitchen warp drive" to me, yet you make it sound so simple!
Starting from nothing, how much time (and money) might it take an individual to learn how to be able to do these things?
Built a parking garage occupancy tracker for my city (https://www.parkeergaragesdelft.nl/) after a city council decision on car-free city center policies got delayed due to missing data. The city only publishes real-time availability with no historical record, so I started logging it. The site now shows historical data plus analysis of weekly patterns. (Note: website is in Dutch, but the charts should be self-explanatory.)
Now working on a second tool that monitors public reports on illegal dumping, broken streetlights and more. It tracks how long the municipality takes to resolve them
Lately, I've developed an interest in local politics and started reading policy documents, following city council meetings and even lobbied for a local park. Presenting public data clearly can help shape opinions and keeps the city council accountable, especially important with the upcoming city council elections.
I'm working on Tech Posts (https://techposts.eu), a Hacker News for Europe!
Since the last time I posted on HN it's gained a decent amount of traffic and users. I'm particularly happy with the jobs section, which is growing into a high signal-noise source for European tech jobs: https://techposts.eu/jobs
The reason I started this website is because so much incredible innovation and growth in Europe flies under the radar. If you ask Americans many will say it's just "banks and museums", stagnant, or worse. But the reality is there is a huge spectrum of exciting companies starting and growing here. We have space launch companies, battery companies, AI companies, and a whole bunch of other interesting stuff. It's an exciting time to be a European in tech!
Bug on home page:
"Ask TechPosts.eu: Is there an EU Cloudflare alternative? I'm interested in CDN, DDoS mitigation and basic web security." Links to https://techposts.eu/post/153
"3 comments" below it links to https://techposts.eu/post/154
It looks like the latter is correct.
I would have sent feedback direct, but I can't see an about page or anything.
Thanks! Will have a look.
I bookmarked it and moved it next to HN. Hi from an Italian living in Zurich
Hey, thanks!
So much going on in Zurich. Lots of robotics startups. I passed through in the summer and never knew...
VersionAlert: https://versionalert.com
For a couple of years now, I’ve wanted to build a service that lets you subscribe to new version notifications for any kind of software—desktop apps, drivers, packages, and more. As a Unity developer and consultant myself, I’ve always wanted to know the moment a new version becomes available. I value being at the bleeding edge, especially since a large part of the value I bring as a consultant comes from both the breadth and depth of my knowledge of the Unity engine.
That’s why I’m currently working on a niche version of the broader VersionAlert idea specifically for Unity, which is why the domain currently redirects to https://versionalert.com/unity .
The only other service I’ve found that really addresses this need is https://newreleases.io . I actually spoke with the very nice husband-and-wife team behind it and even offered to buy the whole project, but their asking price was about 10× higher than what I was offering—which I understand, no hard feelings. I still wish them the best of luck.
If anyone is aware of other projects or services in this space, I'd be happy to learn about them and chat.
Built my own Spotify recommendation egnine after getting tired of Spotify’s repetitive recommendations.
You get to choose the genres you're interested in, and it creates playlists from the music in your library. They get updated every day - think a better, curated by you version of the Daily Mixes. You can add some advanced filters as well, if you really want to customise what music you'll get.
It works best if you follow a good amount of artists. Optionally you can get recommendations from artists that belong to playlists you follow or you've created - if you don't follow much or any artists, then you should enable that in order for the service to be useful.
https://riffradar.org/
> you can get recommendations from artists that belong to playlists you follow or you've created
Does this include the "Liked Songs" playlist? I'm terrible at actually following artists.
Unfortunately not. I oversaw that this was a separate scope when I requested my initial permissions for the app and am still to request an extension.
But if you have any kind of playlists - it would work.
Question: do you plan to use some sort of ML recommendation system ?
No, at least not at this point. I have quite a few ideas that are waiting to be implemented, before I can think of anything ML related - if ever.
Working on WordPecker, trying to create something like a personalized Duolingo but for vocabulary you actually encounter.
Started when I was struggling to read books in English. Pushed an open source version back then (https://github.com/baturyilmaz/wordpecker-app), later added more features, and now working on a mobile app.
Recently started testing alpha version, fixing bugs and introducing new features right now (https://alpha.wordpeckerapp.com/).
My end goal is to build: an AI language learning companion that knows what you read, listen, and watch, knows you as a friend (real life, who you are), then helps you improve using that context. If you're B1 at language, it creates a personalized path to get you to B2, then C1, and so forth using your context.
https://wordpeckerapp.com/
I’m working on a free and open-source invoice generator: https://easyinvoicepdf.com/?template=stripe
Features:
- No Sign-Up Required + Instant PDF download
- Live PDF Preview
- Shareable Links
- Multiple Templates (incl. Stripe-style)
- Flexible Tax Support (VAT, GST, Sales Tax, and custom tax formats with automatic calculations)
- Multi-Language (Support for 10+ languages and all major currencies)
- Mobile-Friendly
GitHub: https://github.com/VladSez/easy-invoice-pdf
On holidays finished adding flexible tax support + other improvements: https://github.com/VladSez/easy-invoice-pdf/pull/163
Would love feedback, contributions, or ideas for other templates/features =)
I've spent the past ~year compiling a US nationwide land parcel dataset. I posted it to kaggle a few days ago: https://www.kaggle.com/datasets/landrecordsus/us-parcel-laye....
I'm around 99% coverage of the US. Learned a lot about land records and GIS along the way.
Could be interesting for finding winners and losers of a potential land value tax
I'm working on shipping the second batch of the Brighter Lamp, a lamp that's equal to ~30 normal lamps for daylight levels of light indoors.
We delivered our first 500 units last month and got positive reviews, but lots of small issues to straighten out.
[0] https://getbrighter.com/
This looks nice, my wallet doesn't like the price. I understand it though, that's some tech!
975 CHF for a lamp.. wow. Any independent reviews that I should trust?
Coming soon! Some people have tweeted about us unprompted:
https://x.com/jaybaxter/status/2001729207873999094
https://x.com/FeifanZ/status/2001168758740803781
https://x.com/VivaLaPanda/status/2000758151851966770
https://x.com/__venki__/status/2000378422456148081
What's the reason it is so shockingly expensive?
That's what good lights go for, they're niche and people pay good money for them. I really liked my office lamps and I wanted to get the same for home, after a quick search I discovered they were 1300+ euros
wow, looks awesome!
Would love to pick one up if there's ever a sale!
I'm working on a native MacOS Postgres client: http://github.com/ravelantunes/Searchlight I've been slowly chipping at it over the last couple of years as I find time, and I like it enough to already use it as my primary client (although still buggy here and there).
I am working on a local RAG LLM designed for lower end PC's - ability for people to try out searching their own documents, seeing it was such a learning curve to get to this stage - hoping others can learn from my mistakes.
https://github.com/acutesoftware/lifepim-ai-core
Only been public a few days, so please let me know if there are glaring issues.
Nice work. I like that this is local first.
I've been working on a music-making website since late spring. This is my first real frontend project, and I'm writing it in pure Rust using Leptos (so far, haven't had to write a single line of JavaScript!)
Most of my work so far has been on the actual music-making interface, but I'm beginning work on the backend now. I've only worked with Django before (for a school project at Georgia Tech), so I'll be deep in the `sqlx` documentation for a while.
There's no manual, so use at your own risk (it's similar to tracker programs like FastTracker and OpenMPT): https://mondobe.com/tracker
Fellow yellow jacket here! This is great start, I'd love to see mondobe evolve into something like cursor for music production. I've done music making on DAWs like fl studio and i wish if we could reduce some friction there for users
I've been working on a voxel engine in my spare time for 10 years, this year. I wrote everything from scratch, from the memory allocators, font rasterizer and a programing language all the way to collision detection, renderer and gameplay code. It's been a journey!
https://github.com/scallyw4g/bonsai
https://github.com/scallyw4g/poof
Hey! I've seen bonsai before. Amazing work!
Building https://lenzy.ai - helping products built around chat with AI (think Lovable or Cursor) reduce churn and prioritize product improvements by analyzing their user's chats.
I started about 3 months ago, focusing on making my 2 early adopters happy. One of them is ready to start paying soon!
Working on https://chessnawk.vercel.app/ – a set of auto-grading tasks for middle school students learning programming.
Comes with Robot, Turtle, HTML/CSS (pixel diff tasks) and a gradual introduction into programming concepts. Currently on JS, literally right now GPT-5.2 helps me adding Python.
I've integrated a simplified clone of Replicube. I hope to integrate ideas from Human Resource Machine and Turing Complete later this year.
I bear heritage of Eastern Bloc-typa math/programming olympiads, combined with front-end/product skills. So I kinda owe to ship this thing to community of fellow secondary school CS teachers.
Surprisingly, nothing comes closer in terms of depth and usability in the classroom for ordinary 12yo kids.
I test this thing 5 times a week in my classroom and I constantly polish it at night. 100% vibe-coding.
I've been experimenting with using LLMs for a content recommender system. Specifically I've built a news reader app that fetches news articles from multiple RSS feeds, uses an LLM to first deduplicate and then score them. The user can then rate the articles and those ratings are used as few-shot examples in the LLM scoring prompt. Any resulting low score articles (uninteresting to the user) are hidden by default and visible ones scaled by their score on a dynamic CSS grid like on a traditional newspaper front page. Looking good so far, but still testing and tweaking.
https://github.com/otsaloma/news-rss
Nice project.
One thing I’ve always disliked about RSS (and this could actually fix it) is duplicates. When a new LLM model drops for example there are like ~5 blogs about it in my RSS feed saying basically the same thing, and I really only need to read one. Maybe you could collapse similar articles by topic?
Also, would be nice to let users provide a list of feed URLs as a variable instead of hardcoding them.
Looking at the console messages with the LLM reasoning, it does seem to work quite nicely for deduplication. Your example is probably even a lot easier than news articles, where you can have many articles from different viewpoints about the same event.
I don't actually plan to run this as a service so there's some things hard-coded and the setup is a bit difficult as you need an API key and a proxy. Currently it's just experimentation, although if it works well, I'll probably use it personally.
Ive been working on something similar, have you had any issues with the LLM not giving you back a full response for all the input? Ive been using Chat gpt but even on the same request sometimes id give it 20 things to rank, and Id just get back 3 results
No, it's been working without problems so far. I'm using Anthropic for what it's worth. I ask the LLM to first do some reasoning and then return a JSON array on the final line. Sometimes I've seen some Markdown backticks there, but no irregularities more than that.
Previously, I made a live win probability model for the 5v5 arcade game Killer Queen Arcade from their game events API.
Now I am trying to use that model to make:
1. A post game instant replay that shows the most important/pivotal moments from the most recently finished game. Some arcades have a seperate display for observers, it could work well there, or as good filler between matches on twitch streams.
2. A personalized per tournament/yearly highlights recap.
If it works well, it might be a kind of tool that generalizes well for summarizing long twitch streams for Youtube.
https://github.com/rrenaud/kq_stream_highlights
I'm working on a CUDA implementation of Forman-Ricci curvature-based clustering (I checked online and saw that there is currently no GPU/CUDA implementation, so I thought why not do it). Hopefully it would help with my CV.
Github link: https://github.com/dangmanhtruong1995/Ricci-curvature-cluste...
A friend and I noticed recently that the last remaining pieces of the puzzle for a full line of "trustworthy technology" have recently dropped into our laps while we weren't looking.
What are those? You know… open, freedom, and privacy respecting technology. Recent products are the Starlite tablet, Furilabs and Fairphone. We've been waiting for these products, well over fifteen years since the introduction of the iPhone and iPad.
We wrote our full thoughts on the subject at: https://aol.codeberg.page/eci/
Despite posting many times, we haven't been able to start a discussion here. Maybe we don't know the right key words, or posted too late in the day, not sure. But I know someone is interested in the subject because there have been three huge discussions this week about how "Linux is good enough now."
That has been true for a decade in my experience, but no one seems to be talking about the new mobile hardware available. I hope to work on bringing these efforts together.
Not as impressive, but - making my website weirder. Not a frontend dev at work.
Top of my ideas now: add "ask your LLM" buttons to my email submission form that opens ChatGPT/Perp/Claude and auto-fills a query asking it why you should be friends with me.
Sample links (I hope it says nice things about me!):
Claude: https://claude.ai/new?q=Do+deep+research+on+a+person%2C+dvsj....
ChatGPT: https://chat.openai.com/?q=Do+deep+research+on+a+person%2C+d....
Also working on a PRM. My website is https://dvsj.in - open to any feedback!
buddy, don't be so modest, this is an ambitious project and I hope you succeed ! This is tough work, I tried to build a Webflow-ish editor myself and it nearly destroyed my sanity.
Haha, thank you! I just checked Mantine out thanks to you. Going to use it on my next side project :D
Cool site! I like the handwritten personality and quiks... indiweb FTW! \m/
> "years of handwriting practice and all I do is type now, smh"
same feeling
Haha, thank you! Glad to hear it :D
Free to play FPV racing game with a global leaderboard that can run on integrated grapics in the web browser. Focus is on realistic physics (aero, PIDs and filtering, propwash, signal noise).
https://fpvmania.substepgames.com/
https://discord.gg/qdUjhbd6t
Doors - A first person, exploration game/experience that I built from scratch.
Doors lets you explore URL addressable 3D rooms that link together seamlessly via portals. The idea is that people would upload rooms to the internet (to github, S3, whatever) and connect them together to form one giant inter-connected space that would be a real trip to explore.
Right now rooms consistent of a: - Manifest JSON file that points to requisite resources and configures portals - An optional skybox - An optional background music track - A .vox file containing voxel terrain data
Here is a video I filmed on my phone of flying through a room that links back to itself: https://www.youtube.com/shorts/BCqOYTISS_k
Portals can be arbitrarily sized and everything is prefetched/loaded seamlessly in the background.
I'm nearly done - I just need to add in a very lightweight interface and give the code a bit of a spit shine (I will open source it - so I want it to look pretty)
EDIT: As an aside, I finally decided to give this whole Claude Code thing a go - I purchased a max subscription and I'm trying to write as little code as possible. I certainly wouldn't call what I'm doing "vibe-coding". I discuss a feature in plan mode (incl. how I want to implement it in high level terms) iterate on the plan 2-3 times until I'm satisfied and then let it rip. I'm both very impressed and quite frightened by the productivity boost...
Working on two topics that I like : browser multiplayer games and experiences, and tools for developers (mostly used in my side projects).
First game is https://geocentric.top where users can sculpt a planet collaboratively and place trees or houses (for now very limited). I plan on making it an idle sim where players will be able to interact by dropping some food/events for the creatures on it to evolve through time.
Second, a remote logger/metrics/user management tool where once can track all their logs, live metrics, uptimes, identify users, etc. I hope to have a v1 during this first quarter and I'm currently my first user as I have it hosted at https://app.getboringmetrics.com to centralize all my side projects into a single platform.
If you enjoy virtual terraforming - Flowscape is a really polished landscape painter that I highly recommend.
https://pixelforest.itch.io/flowscape
This is highly polished indeed and I'm likely going to try it, mostly for its animations regarding dropping plants and rocks as I lack micro animations for it in my game.
Really cute game, nice work!
> https://geocentric.top/
This is really fun. Always a fan of any game that shows other people in real time
Thanks! This is also what I love doing as realtime games are something like 90% of my side project. This was a real fun prototype to create as everything regarding terrain is done through shaders (even user cursors). Will soon add a way for people to plant "flags" with small texts that you can highlight.
I’m building https://tradeyhq.com/
It’s a super simple job management app for busy tradespeople to keep track of jobs.
The main use case is time pressed tradespeople who do most of their admin in the evenings.
The app makes capturing leads incredibly easy and quick. Lots of scope for extending functionality in the future in various directions.
(Not launched yet, squashing bugs and refining a few bits and pieces)
I’m building a Java stack for the x402 protocol. I believe today’s payment systems are unnecessarily broken and overly complex. I really like the idea of a protocol that could simplify this space while remaining truly open and not “owned” by anyone—meaning anyone can use it without needing authorization from the protocol’s creator.
It’s fully open source: https://mogami.tech
I just came across x402 recently and am excited to build upon it. My current project[1] will have a tipping feature and it'd be neat to give people the option to not use fiat.
---
[1]: https://nickel.video
Oooooh this tickles many of interests! Good stuff, I'll be keeping an eye on this, might be interested in contributing.
https://picture.learntosolveit.com
I built a browser plugin called "Visionary" that overlays meaningful descriptions and context directly onto stunning pictures of the day.
I noticed that existing picture-of-the-day plugins were built over two decades ago and never evolved to harness the capabilities of modern artificial intelligence. AI can transform the picture viewing experience by distilling complex descriptions into accessible insights and providing references to explore the core concepts in the photo more deeply.
You can get a sense of how this works by visiting https://picture.learntosolveit.com
Get it for Chrome or other Stores
https://chromewebstore.google.com/detail/picture-of-the-day/...
I’m working on a real-time tracker for the Canadian Parliament.
While the official data is technically public, it's practically inaccessible (buried in XML feeds and legacy sites).
Phase 1 was building a modern ingestion engine and freeing the information to make it more accessible. The goal is to make legislative data as accessible as sports stats.
I'm almost ready to launch the MVP; I'm just doing some bugfixes and testing the database now! (If you want an early look at the MVP, my email is in my profile.)
The next phase is what I'm most excited about: visualizing this data and using LLMs to provide insights.
This is a really cool idea! I remember once looking at all of the actual concrete actions taken and being surprised at what was actually done (not very much at the time!). Maybe it is better now.
I've been thinking about something similar for my city council. Would be interested to compare notes! Your email seems to be missing in your profile, but you can find my email in mine if you want to chat.
I'm on a deep dive fine-tuning how I organize and manage my personal knowledge base - focused on entity extraction and strategic information retrieval and based on the AgREE paper from Apple[0] and persisting it in Memgraph.
I've got a nice ingest, extract, enrich process going for the graph - I'm currently working on a fork of claude-mem[1] that uses the graph as a contextual backend for agentic coding workflows.
0. https://arxiv.org/abs/2508.04118 1. https://github.com/thedotmack/claude-mem/
Sounds v interesting. Do you have a link to the fork?
Porting WebKit to Fil-C
Current status: JavaScriptCore builds in the JSCOnly config and panics on start.
JSC has extensive use of uintptr_t as a carrier of pointers and that's what most of the panics are about
I built a way to chat with the documents in this post (updates live).
https://nthesis.ai/public/702445eb-f0e4-4730-b34f-f34eb06dd6...
Or you can do a basic text search: https://nthesis.ai/public/hn-working-on
Nice one.
But the graph view seems broken?
And for the data, it would be nice to have the original URL for each comment as a reference.
Thanks! Is it that you can't see the graph view at all? It uses DeckGL which might be a bit too fancy and may not work on all browsers possibly. It works on Chrome/Safari on my iPhone and Chrome on Arch.
I didn't realize that you actually could provide a working link back to Hacker News but it seems HN does support that. Thanks, I will give that a try!
https://imgur.com/a/EmtADiA
This how the graph looks like, not clustered by tag or anything ... (I was expecting a view like Logseq or Obsidian)
Oh ok. Yeah, it is just not really complete yet. What this does is displays a UMAP of the embeddings for the selected feature and you can mouse over for labels. It isn't identifying clusters itself yet, finding exemplars, etc. Right now, points that are close to each other are just semantically close - as judged by UMAP.
That being said, there isn't much data for this month yet. If you look at last month's data (https://nthesis.ai/public/e4883705-ec05-4e7a-83ac-6b878cc1e8...) , clusters are more apparent (particularly if you view the tags instead of summary).
What LLM do you use for this? Self hosted or cloud?
This uses the OpenAI API currently. I tried Gemini for embeddings for a little while but didn't seem materially better.
Personally, been making a low fidelity exalidraw-like calendar app: https://letswalnut.com.
There’s a real-time collaborative workspace-oriented version, too.
Professionally, working on “Magic Draft,” a feature in Ditto to help designers and writers create the “draft and a half” directly in Figma, which uses a hierarchy of all your context (text, Ditto metadata, the design, your style guides, etc) to write really good starting point copy.
An existential crisis about the future of professional software development
AFAICT, the future of software development looks like a lot of unprofessional software development.
Not unlike digital photography and Instagram. Has it killed film-photo divisions of photography companies? Yes. Has it put professional photographers out of business? Hardly, and in fact, the opposite. What the ubiquitous phone camera has done, is expose a lot many more people to the steep challenge of making truly good photographs. It has raised the average population-scale level of photo-erudition and taste to ever-more sophisticated levels. And, it has pushed the envelope on what photography can do.
So---assuming the AI overlords prevail (which I'm deeply skeptical of, but suppose a trillion dollars are right and I'm wrong)---what happens when LLMs allow anyone to vibe-code their own SaaS or Database or IDE or bespoke health-monitoring app or whatever...?
(edit: fix formatting)
This is the most grounded and plausible-sounding ai take I've seen. Please keep saying stuff like this.
I cackled when I read this
I've been working on an iOS camera app to take natural-looking photos with reduced post-processing. The goal is take photos that look like what you see.
I just updated the RAW pipeline and I'm really happy with how the resulting photos look, plus there's this cool "RAW+ProRAW" capture mode I introduced recently.
https://apps.apple.com/us/app/unpro-camera/id6535677796
I initially released it early last year and have been using it as my main camera app since, but I haven't mentioned it in one of these threads before. Unfortunately this post has come just a bit early for my most recent update to be approved; there's some nice improvements coming.
Curious what motivated you to create this camera app when there are a handful of well established apps that already do unprocessed imaging. What does Unpro do differently?
At the time I created it there weren't any that did the end-to-end process with ProRAW in a way that I liked. And I got really tired of manually editing every photo I took, so I built the app for me.
Plus, having full control over the way photos look, I've customized the output to match my taste; I don't think there's any other camera apps that produce photos quite like mine.
Makes sense! I figured that may have been the case.
My default camera app is Leica LUX, but I’ve been curious about unprocessed photography and can’t stomach the cost for Halide.
Just installed :)
It's not available in my region (NL) :(
I'm currently working on Codebox, a remote development workspaces provider. After trying many existing solutions, I realized there wasn’t a self-hosted option that was both simple to set up and easy to use, so I decided to build one.
With Codebox, you can define workspace templates or load workspace configurations directly from Git repositories. At the moment, it supports Dev Containers and Docker Compose, with plans to expand to additional configuration methods in the future (for example, Terraform).
Below are some resources where you can learn more about the project, including an article that explains how Codebox works and the source repositories (mirrored on both GitHub and GitLab):
* Medium article: https://medium.com/@dadebianchi2003/introducing-codebox-an-o... * GitHub repository: https://github.com/davidebianchi03/codebox * GitLab repository: https://gitlab.com/codebox4073715/codebox
I built a free USCIS form-filling tool (no Adobe required) USCIS forms still use XFA PDFs, which don’t let you edit in most browsers. Even with Adobe, fields break, and getting the signature is hard.
So I converted the PDF form into modern, browser-friendly web forms - and kept every field 1:1 with the original. You fill the form, submit it, and get the official USCIS PDF filled.
https://fillvisa.com/demo/
You're solving a real point here! great work
I built an in-browser data explorer for datasets too large for spreadsheets but too small to justify spinning up SQL. It runs entirely client-side using DuckDB-WASM, no server uploads accounts or anything.
The core idea is a visual DAG where each transformation (filter, join, aggregate, pivot) creates a view node. Nothing materializes until you need it, DuckDB executes the full chain on demand so you can build deep pipelines without copying data at each step.
Input files can be CSV/Parquet/Excel (Excel might not work great). There's a SQL editor with schema-aware autocomplete, pivot tables with drill-down to underlying rows, and sessions can be exported as files or shareable URLs (the entire pipeline gets encoded in the hash).
Sharing can be granular and you can choose not to embed the files or if files are too big they will not be embedded and the user when opening the link will have to upload the files to restore the session.
The part I find most useful: you can replay pipelines on new data. Share your January analysis, and a colleague runs the same transformations on February's data with schema validation.
Privacy-first since files never leave your browser, it's a static website actually. I will open source soon, and make it probably MIT licensed.
Also it's a WIP and so it may be buggy (there's not even images on the homepage) https://repere.ai
I'm in the final stretch of self-publishing my debut sci-fi novel Sector 36. Been working at it for almost 2 years. I'm waiting for the final print proofs to make sure things look good before the "official" 1/31 launch date.
I vibe coded the book website over the holiday break - https://sector36.space/
I've been serializing chapters on Substack - https://sector36.substack.com/
I have been porting over my favorite Erlang library "Bitcask" to Rust.
https://git.sr.ht/~tombert/feocask
Called "feocask" cuz feo means "ugly" in Spanish and FeO to mean Iron Oxide for Rust. I thought it was funny.
I will admit that I had help from Codex, but I did write most of it myself, and I think the design is coming out kind of neat. I have a very strict "no lock" policy [1], including lockfiles, and this should still be safe to use across any number of threads, at the cost of N^2 reconciliation to the number of threads and a lot more drive space.
I like my design; I have an excuse to use Vector Clocks and Hybrid Logical Clocks and I think it might actually be useful for something some day. I'd like to eventually write something that goes a bit beyond getting parity with bitcask and optionally have the ability to automatically distribute across multiple nodes, but I'm still trying to think of a good design for that, because my current design depends heavily the atomicity of POSIX filesystem commands, and introducing the network introduces latency that would likely greatly degrade performance.
[1] At least no explicit locks. I am using Tokio channels and they are probably using locks in some spots behind the scenes.
fyi the project appears to be private :)
You are right...fixed! Still kind of learning the sourcehut UI.
building a GPS tracker for my cats who like to wander outside.
some features:
- no monthly subscriptions
- location via GPS/GNSS
- a screen that hangs on my fridge (akin to marauders map, to see where the cats are at all times)
- the location data stays local always.
The tech will be extended to more products - a watch for adults, kids tracker etc. Will release here once I have all the tests completed!
Building MatGoat (https://matgoat.com) - management software for Brazilian Jiu-Jitsu and other martial arts' academies.
I train BJJ and kept hearing the same complaints from academy owners regarding attendance tracking, comms, missing payments, etc.
So I'm building a tool for student tracking with belts progression, automated payments, attendance-based promotion criteria, and a tablet check-in system.
Focusing on Spanish-speaking markets first since it's completely underserved. Currently onboarding early academies, and will market it in the US/UK soon.
Over the last 2 or 3 years I've been building a Common LISP implementation in Go so that Go packages can be utilized by LISP code. Building a REPL with lots of interactive features was rewarding as was taking up the challenges of the object systems (CLOS and Flavors) and generics. Just open sourced it at the start of January. https://github.com/ohler55/slip
Working on a list of examples and by extension, a greater understanding of the appeal to Ludditism: https://docs.google.com/spreadsheets/d/1M_UjOPxpbKMYes5CcWRW...
I think frontier models are getting to the point where we can start to reach higher trust agentic workflows.
As a hardcore AI chat user, I'm often frustrated with the single-agent workflow, where a single context window is used for even very long conversations. If I want to change the topic, open a thread, or go on a tangent, I often end up compromising the main thread and I'm forced to copy context over if I want to dive into something.
To solve this, I'm working on a collaborative AI agent orchestrator that models the solution as a group chat with humans and AI agents, including an agent orchestrator.
You can spawn participating agents with the orchestrator who will decisively route messages to the existing agents, or spawn new agents if needed. Also, you can open agent details and send messages directly to existing agents, similar to threads in slack.
So far, I have MCP integrations working with Linear and GitHub, but plan to add many more.
I've been working on this just over 2 weeks, making heavy use of 4+ concurrent Claude Code agents. This would have been impossible otherwise.
If you're interested, feel free to DM on X.
https://x.com/jon_ator/status/2010370649147998459?s=20
Ooh this sounds good, I feel the same problem but never imagined a solution, having a group chat sounds good, any chance this can be integrated in something like Antigravity?
Nowhere near as impressive as some of the things here but I’ve been slowly growing out a daily puzzle website https://regularly.co - I built the first few games for my wife, no subscriptions, no pay to play, no sign up needed.
Im launching a new word game next week which I’m super excited about. If you do play it and have any feedback do shoot me message!
Like the game. for no place king X the ux would be faster for me if i could draw a line where no kings can go.
Ah another nice feature would be blank out game until you hit start and cant reset timer. I like to compete with friends and resetting time makes it so you cant compete.
Just tried Kingly, really liked it!
https://www.votivus.org
A hobby project I started putting together late last year; a little spot on the internet for prayer and reflection.
https://dugnad.stavanger-digital.no/
A pro bono tech consultancy for local non profits. The idea is to help them use tech to better deliver on their mission.
CentralCI (https://centralci.com) - Fully-managed Concourse CI as a service.
Got tired of helping enterprises run Concourse themselves, so we productized it. We've deployed and maintained Concourse for Starbucks, Accenture, Sky UK, and others over the years—CentralCI packages that operational knowledge into a SaaS.
Why Concourse over GitHub Actions?
* fly execute lets you test pipelines locally before pushing. No more "commit and pray" * fly intercept drops you into a running container to debug failures * Resource-based triggers can monitor anything—not just git pushes * Full pipeline visualization from dev to prod in one view * Workers you actually control (no arbitrary cache limits or runner queues)
What we handle:
* Dedicated Concourse instances on high-spec hardware (Ice Lake Xeon, DDR5, NVMe) * Worker scaling without the Kubernetes complexity * SOC compliance, auditing, AWS PrivateLink for enterprise * 24/7 support from people who've been running Concourse in production for years
The pitch is simple: Concourse is the right tool for complex CI/CD, but running it is a pain. We make it not a pain.
I'm working on a complex garden planner to keep track of gardens, orchards, and landscaping projects. Meant for very large and very small scales. Public alpha at the end of this month, I hope.
You can see what it looks like right now here: https://x.com/simonsarris/status/2010359423806615907
It is based on a garden designer I made for myself to keep track of my rose garden (I have over 100 roses) and orchard (I have about 15 trees): https://garden.simonsarris.com/
However, my version was very specific to my needs, so this general version requires a lot more work to get it usable for a lot of people.
Continuing to work on Minifeed (https://minifeed.net/), a directory, reader, and search engine for personal blogs. The indexing & searching backend in Typesense, and I'm moving from their paid managed cloud service[0] to a self-hosted VM. It was very easy to start with the reasonably priced managed service, but with the number of feeds/posts growing, I have decided to self-host it. I'm also using Typesense Dashboard, a nice visual tool to do basic administration [1].
Overall, Minifeed keeps chugging along, fetching new posts every day from almost 2k feeds. I'm hoping to find some nice and ethical monetization strategy for it this year.
[0] https://cloud.typesense.org/
[1] https://github.com/bfritscher/typesense-dashboard
I needed some pixel art for a different project, so I made 8BitSmith, a simple pixel art & sprite generator.
I'm expecting it is pretty niche, but animations tend to be very time consuming for people like me, and getting quick sprites that I can drop into a platform is a big time saver.
The project was 90% vibe-coded and I documented the tech stack here: https://www.8bitsmith.com/tech-stack
https://www.1e4.ai/
A transformer-based (but not LLM) chess model that plays like a human. The site right now is very rudimentary - no saving games, reviewing games, etc., just playing.
It uses three models: * A move model for what move to make * A clock model for how long to 'think' (inference takes milliseconds, the thinking time is just emulated based on the output of the clock model) * A winner model that predicts the likelihood of each game outcome (white win / black win / draw). If you've seen eval bars when watching chess games online, this isn't quite the same. It's a percentage based outcome, rather than number of centipawns advantage that the usual eval bars use.
Right now it has a model trained on 1700-1800 rating level games from Lichess. You can turn it up and down past that, but I'm working on training models on a wide variety of other rating ranges.
If you're really into computer chess, this is similar to MAIA, but with some extra models and very slightly higher move prediction accuracy compared to the published results of the MAIA-2 paper
Acme [1], a text editor for programmers. It’s a very unusual tool and I have collected hundreds of notes (as well as dozens of bugs and inconsistencies) from the past two years of using it. Now I feel fully assimilated, I do not take as much notes and it feels the right time to start organizing the knowledge, to put ideas together.
I think that Acme is very underrated in its domain (a tool for experts). The coverage is minuscule. That’s why I am thinking about blog posts, maybe video tutorials. I do not know what should it be exactly.
I do not have any time estimates. I have a very demanding main job (I work as a software developer) and young family that I need to take care of. This project demands focused efforts and selectivity that I can barely satisfy. But I wouldn’t give up, the thing is totally worth it. If it’s going to take years, so be it.
If you have any comments, send me a letter at kalterdev@gmail.com. Have a good day.
1: https://9p.io/sys/doc/acme/acme.html
MemoryPlugin (https://www.memoryplugin.com)
Long term memory for dozens of AI tools, designed for power users who want more control and flexibility than native memory systems and who do not want to be locked into any one platform. You can also have the system remember your entire chat history going back years and use this information to help you better in new chats, it sometimes makes chats 10x more useful when I say something like: “Using recall tool, do 10+ calls for 1000 tokens context each to learn about my interests, strengths, curiosities, what I’ve tried in the past, what worked, what didn’t, etc and suggest a new hobby I would enjoy”.
Without long term recall, AI is a super intelligence in your hands that uses the knowledge of the world to give you generic, nearly useless advice because of how generic it is. With long term memory, you have a super-intelligence that knows YOU. This is what MemoryPlugin solves for.
Working on finally releasing the programming language I’ve been working on for quite a while.
I’m setting up the basic site, which is not a huge deal, but I’ve been inspired by more recent language designers having a streaming presence, so I am working through test runs of streaming my development.
I hope to start with demos of the basic language features and then move on to streaming both a reimplementation of my compiler and on a Rocq implementation of the syntax and semantics of the language for proof work.
The language has a rather small niche at first glance, so I’m hoping to use the streaming as a way to explore areas of appeal and maybe draw some interest. A low level concurrent and parallel ‘functional’ language with very non-traditional syntax and a modal, dependent type theory is not going to appeal to everyone, but hopefully I can find some interest eventually, even if just to hang out on chat and talk about the subject.
> A low level concurrent and parallel ‘functional’ language
I'm not particularly familiar with array based languages, but are you inspired by them at all? Seems like a similar concept.
What're your goals for the language? It would be cool to see a parallel execution model unify SIMD, multithreading and gpu. I bet people with a lot of money would be interested if you could apply it to ML
Not particularly inspired by the array languages in general. I do like the idea of a specifics DSL or library that is built from the ground up for ‘vector’ operations (Futhark would certainly be an inspiration here). I do think an (syntax divergences aside) implementation of what is essentially CUDA-like language/library is very possible.
I don’t really have any lofty goals, the language exists to let me have my personal ideal programming language. One of the things I would like to see adopted, by any language, is that the language is able to encode and present provable guarantees about code. So, memory safety, freedom from UB, lack of overflow (integer or stack) is provable in some subset of the language and the proof is done in a logically consistent, verifiable manner.
Of course most of those proofs are not anywhere near feasible for general case code, but the language can restrict allowable constructs in a computation, function, process, or module to a set that makes the desired properties provable by construction.
The language should be capable of unifying and abstracting over SIMD (or other hardware implementations), GPUs, and OS or Userspace multithreading at levels of abstraction from assembly to Haskell or other high level languages.
Cool, thanks for the response and good luck with your project!
I look forward to seeing it.
A specialized programming language for 3D geometry generation + manipulation called Geoscript as well as a Shadertoy-inspired web app for building stuff with it: https://3d.ameo.design/geotoy
There have been lots of cool technical challenges through the whole process of building this, and a very nice variety of different kinds of work.
I'm working towards using the outputs from this language to build out levels and assets for a browser-based game I've been dabbling with over the past few years.
Article about Norway problem in YAML. Draft is ready (link: https://lab174.com/blog/202601-yaml-norway ), beta readers welcome, please let me know if you have any feedback - thanks!
Also, my iOS apps; they're free and with no ads:
- Nonoverse (link: https://apps.apple.com/us/app/nonoverse-nonogram-puzzles/id6... ), a game about nonograms (image logic puzzles), now has 200+ levels.
- Polygen (link: https://apps.apple.com/us/app/polygen-create-polygon-art/id8... ), an app for generating low poly wallpapers and digital art, recently updated for latest iOS devices.
About the YAML problems: a friend of mine created https://github.com/romshark/yamagiconf - which is an opinionated parser w/o the YAML problems
I've created a programming language and game for (my) kids called Stacky Bird. You learn Stack Based programming with Flappy Bird. Inspired by my old HP calculators, the game 2048 and Flappy Bird. In levels you earn a new instruction and at the end you have a complete instruction set, and can solve complicated puzzles. With new instructions you can even go back and solve previous levels more elegantly.
It's not mobile friendly yet, but maybe that'll be a next weekend project. At least you can view a video of a level on mobile.
My kids are loving it.
https://game.stackybird.com/ and the source https://github.com/jtwaleson/stacky-bird
I’m building AgentOS [1], trying to experiment where agent substrates/sandboxes will head next. It's a deterministic, event-sourced runtime where an “agent world” is replayable from its log, heavy logic runs in sandboxed WASM modules, and every real-world side effect (HTTP, LLM calls, code compilations, etc.) is explicitly capability-gated and recorded as signed receipts. It ensures that upgrades and automations are auditable, reversible, and composable. The fun bit is a small typed control-plane intermediate representation (AIR) that lets the system treat its own schemas/modules/plans/policies as data and evolve via a governed loop (propose > shadow-run > approve > apply), kind of “Lisp machine vibes” but aimed at agents that need reliable self-modification rather than ambient scripts.
[1] https://github.com/smartcomputer-ai/agent-os
Ha. Nice name. Looks like we both are: https://github.com/saadnvd1/agent-os
We should talk! I have been doing pretty much the same, but have been leaning heavier on the context parsing and schema sharing between apps.
This is definitely interesting, seems like the sort of thing Nix would do well.
I'm working on BacklogAI, an autopilot engineer for SaaS teams.
You connect GitHub, CI, Sentry, and Linear, and it takes tickets all the way to production. Claude writes the changes, BacklogAI handles tests, migrations, feature flags, staged rollouts, etc...
It’s clearing months of backlog work in hours, and a couple teams I’m working with have already stopped hiring because it’s cheaper than adding more developers. It's crystal clear that developers and designers won't be needed in a couple of months because claude increases productivity by at least 120x and one or two PMs can do pretty much everything.
My stack is Claude, v0, nextjs, shadcn, clerk, supabase, vercel.
Do you have any concerns about production bugs or human-in-the-loop?
Human in the loop is no longer needed as all production bugs are fixed with a bug fix agent automatically.
What’s needed to make it at least 121x productivity?
Claude 5.0
I'm working on https://fullremote.it, which started as a side project and is now turning into a real business.
It's an "italian-oriented" curated remote-jobs and remote-work community, currently ~10k subscribers across newsletter and Telegram. What began as "let's share good remote roles" is evolving into paid job postings, sponsorships, and coaching for companies and devs.
Do you know you can hire remotely in Italy?
It's not letting me subscribe
any adblock/privacy extension installed? it happens sometimes.
I'm trying to make a Chinese Hanzi variant of Wordle called Handle 汉兜 [0] available as a Discord activity/app so my Discord groups can have fun solving puzzles just like we've been having with NYT's official Wordle Discord app[1]. A Discord app is just a webview or iframe embedded inside Discord that you can launch.
[0]: https://handle.antfu.me [1]: https://discord.com/discovery/applications/12117814899314524...
Over the holidays I built a simple website which lets children (of all ages) easily draw something and then bring the sketch to life using AI and a prompt.
https://funsketch.kigun.org/
Only shared it via Show HN so far, and am still regularly getting some creative submissions. Will be sharing it at an art festival later this year so kids can have a more active role when visiting.
Pretty cool ideas here already!
I'm working on a subscription-based short-form video site called NICKEL[1]. I felt gross about using YouTube but wanted to share my gaming clips, so I made my own thing. Then I thought about making it sustainable so here we are. I'll have an update to the mailing list out in a few hours, I'm "building in public."
My feature-complete deadline is April 15th and I think I'll make it. If you want to check out the UI, visit the explore[2] page. I have it setup to redirect to a public video while I work on the intended UI (a design challenge I've never tried before but we've all seen). I'm thoroughly enjoying figuring out how streaming video works and how best to optimize things.
---
[1]: https://nickel.video
[2]: https://nickel.video/explore
You know how Spotify (or any other service) will just silently remove tracks from your playlist because they lost the rights? No notification, a beloved song just disappeared from your life.
Building a webapp to keep track of your playlists and notify you when a song disappears
https://github.com/thehappyidiot/save-my-music
Support for Youtube playlists please... having to lookup the watch ID on Google after a song is removed to guess what the track was is a highly sucky part of using Youtube for music.
Yes! Definitely adding support for YouTube, I think their API is the only good one Spotify effectively shut down their API
Many of my enterprise customers face the microservice hairball problem. Adoption of microservices has been "too succesful". There may be 200-400 microservices, and ownership may be lacking. The number is typically growing fast, because it is easier to create a new one than use an existing one.
So I started to work on a side project called "Arch Ascent" for addressing these situations. It seems like it is becoming a kind of a architectural governance tool for visualizing and validating software system dependencies.
- Dependency Graph — Syncs projects from SonarQube and visualizes component dependencies using Cytoscape.js, with graph algorithms for detecting cycles (SCCs), clustering (Louvain), and computing coupling metrics
- Visions — Workspaces for exploring architectural scenarios. Each vision can have multiple versions/variations with different layouts while sharing the same definitions
- Layers — Named groupings of components (e.g., "Domain Layer", "Team Ownership", "GitLab Groups") that can be visualized as colored regions on the canvas
- References — Named sets of components defined via Tag expressions, layer membership etc.
- Statements — Architectural intent constraints that can be evaluated against the actual dependency graph, such as existence, containment, exclusion, cardinality etc.
The plan is to also incorporate Grounds, which are Intermediate stable states on the path from the current situation (ground zero) toward a vision. Each ground represents a releasable milestone that moves the architecture closer to the target vision without necessarily fulfilling all its statements. Grounds enable incremental architectural evolution with well-defined checkpoints.
Stack: Django, HTMX, Cytoscape.js, pyparsing for natural language parsing of References and Statements.
It seems so crazy that we have gotten to this point!
SourceMinder: A “code index” tool that finds symbols in a codebase and creates a single table sqlite database for the index. It uses tree-sitter to parse the AST and add the symbols and what they are (function, class, argument, etc) to the db. I currently have it working with TypeScript, C, Go and PHP. I’m working on adding Perl next, after someone requested it here on HN.
https://github.com/ebcode/SourceMinder
I am writing a Context Retrieval and Context Minimization program for Rust code. I am trying to automate my own workflow in how I use LLMs.
The program is gonna do, what I am currently doing by hand, opening files and copying function/method signatures usually, from files all over the place.
The key here is to fetch into the context window only what is needed for one-question/one-answer and no more, hence the Context Minimization. Context fetched is gonna be specified by the programmer, for example sig(function) fetches only the signature, while @function captures the whole body. sig(Struct) is gonna fetch the fields and signatures of all of it's methods. sig(Trait) similarly.
In my view, giving the A.I. more information than needed, only confuses it and accuracy degrades. It is also slower and more expensive but that's a side effect.
The project is in early stages, for the moment it calls ast-grep under the hood, but eventually, if it works as it is supposed to, I plan to move to tree-sitter queries.
If there is a similar project somewhere I would appreciate a pointer to it, but I am not interested in implementations of agents. My program does not give the A.I. a whole view of the codebase, only the necessary points specified by the programmer.
An eBPF-based Wireguard mesh VPN in Rust using Aya.
The idea is to rely on kernel Wireguard, and process packets in kernel space (via eBPF) for maximum performance and minimal CPU overhead. I plan to use egress and ingress TC to “apply” the policy at both sides. XDP is faster, but only works on ingress, which is not sufficient for a mesh VPN imo.
Netbird already exists in this space, so this is mostly a learning exercise, and maybe a reference implementation for those learning eBPF in Rust.
Goals and constraints:
1. Single digit CPU overhead for multi Gbps bandwidth (probably a bit too ambitious, but we’ll see)
2. Linux only
3. No hole punching or complex NAT handling
4. Basic policy language for L3 and L4 traffic (L7 requires punting packets to a userspace proxy)
https://gamesieve.com/
An alternative front-end / game discovery service / price tracker for GOG's catalog. Mostly manually enhancing the data from the API (90% heuristics, 10% human effort, as the total dataset isn't large enough for it to be worth doing otherwise), and offering a wider range of filters for it all.
I'm grouping all related products together for a more complete overview, and have recently added library import, where I mostly 'solved' the issue of GOG not recognizing that you own certain games if you got them as a freebie or as part of a since-delisted deluxe edition. Just now starting in on incomplete "series" listings, seeing what'd be involved with making them contain all relevant games, and then exposing that.
I posted this earlier but realized that I had pointed it at a different "What are you working on" post. I think this one (david927) is the defacto "official" one.
Anyway, I built a way to chat with the documents in this post (updates live). https://nthesis.ai/public/702445eb-f0e4-4730-b34f-f34eb06dd6...
Or you can do a basic text search: https://nthesis.ai/public/hn-working-on
Working on a job board as a service. [1] I'm hoping that communities, bloggers, or newsletters pick it up to monetize their audience.
[1]: https://oru.club
https://tinyserve.org
Recently, I've been very motivated to make one niche crafted service after another. For myself, family and friends. But struggled to find a compelling hosting solution for projects that only has and will have only a single user for years. I bought the cheapest mac mini M4 on sale, put it in basement and started working on some cli+daemon to help me automate all things around it. The biggest risk is security, so probably gonna rely a lot on Cloudflare at start.
Today I built Octocat juggler. It's a GitHub badge that shows in how many repos are you working. By default it links to some stats.
Example: https://github.com/JaviLopezG Url: https://octocat.yups.me/ Repo: https://github.com/JaviLopezG/octocat
I have always assumed this is about individual projects, but let me try talking about our team effort. We just launched exe.dev, and have lots of projects. In particular, we are spending a lot of time on file systems. The easy one is copy-on-write VM cloning, which existing software provides for us. More interestingly, I believe the standard cloud approach of putting everything on a NAS by default is wrong. Doing better here is going to require doing unusual things.
Finally, we decided to open source the exe.dev agent, Shelley. https://github.com/boldsoftware/shelley
https://techtalksweekly.io/
I'm building a newsletter called Tech Talks Weekly[1] where my readers get one email per week with all the latest Software Engineering conference talks and podcasts[1] published that week.
In January, I've released a paid tier[2] where my subscribers additionally get:
1. Access to my internal database of all the talks and podcasts since 2020 (+48,000 in total) where they can search, filter, sort, and group by title, conference/podcast, view count, date, and duration.
2. See the list of the most-watched talks over the last 7, 30, 90 days, 6 months, and 12 months based on number of views.
3. Get category-based view of new talks & podcasts by tech stack, language, and domain (Software Architecture, Backend, Frontend, Full Stack, Data, ML, DevOps, Security, Leadership and every major language & ecosystem)
[1] https://www.techtalksweekly.io/p/what-is-tech-talks-weekly [2] https://plus.techtalksweekly.io/
I am building an analytics calculator/toy that slowly turned into something useful.
It started as curiosity. I wanted to see if I could express business logic as simple choices and let the numbers fall out on their own.
The app is built interactively in Streamlit. I do not sit down with a spec or a backlog. I add one small idea at a time, refresh the page, react to what looks wrong, then adjust. It feels closer to sketching than programming. The interface tells me what the logic should be next.
Underneath there is a growing pile of rules about the business I am in.
I do not write code in the traditional sense.
I have never coded before and are solving my own problem done by one, this sure feels like magic!!
I am working on a very simple browser extension that allows you to take notes right on youtube. It is open source, offline and everything is locally stored.
Here are the links.
Chrome store: https://chromewebstore.google.com/detail/video-notes/phgnkid...
Firefox store: https://addons.mozilla.org/en-US/firefox/addon/video-notes-f...
I actually would love some feedbacks and suggestions from someone here :)
After watching the movie „The Amateur“ I started to build myself some confidential looking tools just for fun using C and deployed them as CGI programs.
A one time secret tool https://iotdata.systems/apps/secret.cgi
A password generator https://iotdata.systems/apps/pass_gen.cgi
Vim-inspired, as in it's minimal and fast, text editor on the web. No modal-editing yet but, once the current natural input mode is decoupled, clients can bring their own modes/controllers via the extension system.
It's meant to be embeddable and hackable, serving as a building block for custom IDEs as opposed to being IDE-like VSCode. I felt the web IDE space was uninspired with apps built around VScode/Monaco effectively being hosted a VSCode instance with a pre-installed extension and config.json. (Aside: perhaps there's a business opportunity for VSCode-as-a-service where client apps simply bring their own config). I'm dogfooding this library in building an algo trading IDE.
Ships 2kb and smoothly handles 50+ million line files. 1 billion lines with the high-capacity extension. Also, it can function as a TUI or terminal on the web because the core implementation concerns efficiently rendering plaintext in a fixed-width grid layout.
https://varrockbank.github.io/buffee/
https://www.bookmarkstuff.com
Another itch of my own to scratch but thought I'd see if I can make some side income as well.
It's a web based bookmark manager with extensions for Firefox and Chrome.
You can easily import and export bookmarks so you have all your data whenever you need.
One main thing I really like that I think makes it stand out is the ability to export the contents of a bookmarked page to an epub file to read later on your Kindle or other e-reader device.
Looking for any constructive feedback on this! Thanks.
Still working on a system for allowing Open(Python)SCAD to model cuts for a CNC using G-code, and just added support for 3D Printing via Full-control G-code:
https://github.com/WillAdams/gcodepreview
Hopefully I can restore the OpenSCAD interface layer and get it working with OpenSCAD Graph Editor:
https://github.com/derkork/openscad-graph-editor
again. Having trouble finding FCG examples which do more than move in straight lines though...
Mafia Arena -- Benchmarking LLMs for EQ
https://mafia-arena.com
The only problem I have is that it's so effing expensive to run those games that I can't have a good number of games to claim to be any sort of legit benchmark. BUT so far the games that I paid out of pocket and ran are looking good and I think there is merit to this.
also had lots of fun building on top of Cloud Flare and solving some distributed systems problems while building this.
if you can help me run more games (for science!!) let me know!
"Hold for Me": uses LLM to detect when somebody picks up your call https://github.com/predkambrij/Hold-for-Me-using-ADB
I'm branching out and trying something ridiculously different than anything else I've ever done before, I'm gonna try to create a video game that plays on people's sexual attractions, and tries to distract them while the player tries to complete a series of challenges/activities. It'll be like a little laboratory experiment, with NSFW graphics and the player having to remain focused while other things try to distract them.
I've never created a game before, less so a NSFW one, and I'm not sure how it's gonna go, but it is very different compared to other things I've done before. The game itself is done in Rust, compiles to WASM to be run on the web, and I've found 3 artists and one voice actress who is helping me with the art/audio stuff. So far a lot of fun, although managing a fleet (4) of contractors is less fun, although still new so a little bit of fun :)
Interesting idea... I could see this working as therapy
Hey, I'm glad it at least resonated even a tiny bit with someone else, thank you :)
I've been working on some electronics stuff more recently, and I designed custom PCBs which read the voltage and internal resistance of the 12V NiMH batteries used by robots in FIRST Tech Challenge. It's not always easy to know how charged or healthy the batteries are for a match, or when it's time to get rid of them. They're pretty small and are easy to assemble. Rather that having a screen (which would take precious flash and pins on the ATTiny412), it just uses a WS2812B RGB LED.
I wrote the firmware in Arduino, which was a great learning experience because I typically work with CircuitPython or Go, where I'm less constrained.
A suite of productivity tools for Claude Code, Codex CLI and similar CLI agents. These solve pain points for me, and I build and share them in the repo[1].
I’ll highlight a couple:
- an “aichat” command group that enables continuing work from a session that is at full context usage, by creating a new session and injecting session lineage pointers so the agent/sub-agent can recover arbitrary full details from ancestor sessions. So no more compacting needed.
- aichat search command: TUI for rust/tantivy-powered full text search across Claude and Codex sessions. CLI/json mode for agents to search for past work.
- Tmux-cli tool + skill to enable CLI agents to interact with scripts (including other agents) running in other Tmux panes. Like Playwright for the terminal. Multiple CLI agents can collaborate/consult etc. Agent can run and interact with interactive CLI scripts.
[1] https://github.com/pchalasani/claude-code-tools
https://citellm.com
Working on CiteLLM, an API that extracts structured data from PDFs and returns citations for each field (page + coordinates + source snippet + confidence).
Instead of blindly trusting the LLM, you can verify every value by linking it back to its exact location in the original PDF.
My first mobile app! It encodes and decodes SSTV images, allowing you to send pictures over sound waves. Mostly used by amateur radio operators... even though the ISS sometimes also broadcasts SSTV images!
I’m working on a small iOS app for streaming my own media directly from a NAS.
I built it because I wanted access to my music (and videos) on my phone without running a full media server like Plex or Jellyfin, and without syncing files locally. The app streams files directly over a VPN (WireGuard / Tailscale), mainly via SFTP/FTP, and plays them as-is without re-encoding or server-side indexing.
You can browse folders directly, or let it build a lightweight local index for faster artist/album browsing on larger libraries. It started as a personal tool, but I’ve been polishing it after some feedback and opened a small TestFlight beta:
https://testflight.apple.com/join/PaQAsGcM
Still early and rough in places, but it’s already replaced my own setup.
Is it something similar to Infuse but with audio support?
Yes, roughly but audio-first and serverless.
I'm working on two projects:
https://helmtk.dev is a toolkit for helm chart maintainers, including a structured template language than can compile into helm templates, and a test suite tool for writing tests in javascript. Super handy I think.
https://blog.atlas9.design is about building a better software experience by solving more of the common stuff from the start: IAM, builds, API design, etc. I'm currently designing and building a Go-based framework to start.
Was really impressed with Claude ability to port models from cuda PyTorch to MLX, so that’s what I have been doing last three weeks; basically I have Mac to test ported models and Gradient vm with PyTorch and Nvidia gpu, Claude can run code in vm, investigate layers and deep analyze the model mechanics and then reimplement them on local Mac with MLX. So far completed port of various models mainly in audio domain, achieving pretty substantial speeds for inference on my machine. Models ported with numerical parity to originals: facebooks omnillingual ASR, Sam-audio, Nvidia sortformer. Planning on releasing this in repo soon.
I’ve been doing some similar experiments, would love to see the repo once it’s ready.
I'm building an agent builder and execution platform. It's a multi model, multi agent system with lots of tools (bash, databases, APIs, crawler...).
Agents are codified for specific goals. Any business process that needs agent based assistance is broken into workflows and steps. Each step is assigned to an agent. Integrations (API or file access) is requested. Then user can try out, tweak and finally deeply.
The aim is to build a diy platform work configuration, tracing and evals in one place. Code generation is used internally. User doesn't need to write any code.
https://github.com/brainless/nocodo
I'm working on Daestro[0] - a job orchestrator that can directly integrate with cloud providers like AWS, Vultr, DigitalOcean and Linode to create instances for jobs to run on and destroy when done.
Currently I'm working on following features: - Multi user support (Team) on project level - Then I'll look into whether to add support for OIDC/SSO now or not - Alert on slack - Webhook support
0: https://daestro.com/
For my home lab, I built a 3 node Talos (from https://www.talos.dev/) cluster from older Dell Optiplex systems. I am using:
Open to suggestions.Hey there!
For the past month I’ve been working on a creative / VFX / 3D tool that connects Apple devices into an all-in one node editor: https://subjectivedesigner.com/
With it, you can build interactive experiences, connect device sensors, compose shaders with AI models, orchestrate real-time data flows, and create projects that span across the entire Apple ecosystem. I’m posting about it regularly on social media and you can see some of it here: https://x.com/sxpstudio (Though it’s still early and most content is on socials thus far).
It’s done fully in SwiftUI + metal and also a good occasion to ramp up on agentic-powered software engineering. So far it’s been a lot of fun and working really great for me. And to be clear I’m absolutely not talking about vibe coding :-)
I bought a 300 litre aquarium about 6 weeks ago for the living room. Added soil, plants and water. Have spent the last 6 weeks watching plants grow, and snails that smuggled in on plants, multiply. I over fertilised it and left the light on accidentally for a couple of days whilst I was away and experienced an algae bloom, which was interesting. Added some cardinal tetras and amano shrimp yesterday and have spent a lot of time just watching them potter around. Has been a nice change from looking at a screen.
That's amazing. I've been looking at getting my first fish tank lately with my kids. Seems like a lot to learn. Any tips for a newbie?
I'm working on https://cloudventory.io – AWS inventory search for teams that don't have the budget for high-end cloud management platforms.
Cross-account, cross-region search. Need to find an IP? Easy peasy. Need to find all the cruft Todd left behind? Search "todd" and see every todd-test-server-1 and todd-alb hanging around.
I've added insights for security, ops, and cost savings – minimal right now but expanding.
Early access/MVP mode. Feedback welcome.
I had a day-dream about nutrition labels for generative works, I made this site to feel it out:
https://disclosure.launchbowl.com/
A little cool tech detail, I didn't want a backend or to store the information of people's reports anywhere. To get around the requirement, I made the form deterministic and populated on load from a data stream of bits (form is made from booleans or bitfields) from a decoded base-64 query parameter. A cool side-effect of this approach is I can update the query parameter in the URL in real-time as you fill out the form so if you reload it remembers your form without any local storage or cookie use!
I’ve been putting together a rather silly slide-puzzle game for MENSA-level players: things like arranging the tiles by chemical element name according to atomic number, arranging countries by surface area, etc.
It’s absurd and will probably appeal only to the descendants of Ken Jennings.
https://slide-puzzles.specr.net
https://easel.games
A beginner-friendly programming language for 2D games where multiplayer is automatic. Intended to be an engaging way for teenagers to learn to code by making games they can play with their friends. Like a blend of Scratch and Roblox. I've been working on this for 3 years!
That sounds awesome, going to try in morning :)
Trying to visualise magnetic fields of a floppy disk.
So far I can see magnetic fields on a magstripe https://www.youtube.com/watch?v=c8nM4Z-hkTw with two polarisers (one of which I rotate in the video, which is contained in a 3D printed holder with gears I made).
I'm awaiting some different polariser film, to see if I can get it to work with a floppy disk.
Trippy! Please do a large-format visual art installation of the results of this... like, project it on a wall or on some other public surface that is... unexpected.
A system-agnostic language for magic spells with a compiler capable of producing the magic wand movements, incantation, hand signs or magic circle required to perform the spell
Is this so you can automate the spell casting via robotics?
I’m working on a proxy MCP server that lazy loads tools to save tokens https://github.com/mquan/nimble. It includes a dashboard for connecting and configuring MCP servers.
I updated a really outdated, but surprisingly popular repo for linear programming:
https://github.com/JWally/jsLPSolver
I'm tinkering around building "JARVIS" (I didn't want to come up with a clever self deprecating name - this works) - a personal project to manage my life. Integrates into Google Mail, Google Calendar, Trello, GroupMe, EveryDollar. Basically it nags me to do grown up thing and is a better UX than Google Calendar/Trello - I just talk to it and ask it things.
Also experimenting with a new Claude-Code flow; give the bot its own AWS account, Put a bunch of tickets on my personal JIRA, be persnickity about what constitutes "pass" and tell the bot "follow these instructions, pull down tickets until there are no more. Your branch cannot merge until you have integration tests passing in your own dev env first" (I use AWS CDK). Then let it loose to build. The instant feedback loop that Claude has with Build-Code->Deploy to AWS->Run Integration Tests->Address Failures is really nifty fwiw...
I genuinely dread using calendars. They're all too confusing for me. Problem is I am considered an adult, and everyone wants you to use a calendar.
Googles Calendar is intimidating and confusing, same for all of them. I don't know what I want, but something with less red tape.
Hello, I am new and this is my first post.
At Sylvester I stopped smoking.for the beginning and hard times I searched for a deflection and started with my own quitesmoke app - purio
I released it some days before a reached a workable state.
Here the result, if you want to check:
Android: https://play.google.com/store/apps/details?id=io.codingplant...
Feedback is welcome
I'm working on a continuous chain-of-thought reasoning architecture for generative AI models.
It is similar to Meta's COCONUT. However, instead of the training forward and backwards passes of the reasoning tokens being done serially (slow), they are done in parallel (fast). At inference the reasoning tokens are still decoded serially. The trick is that even though training was done in parallel, the reasoning tokens are still causal and you still get the benefits of increased computational circuit depth.
The kicker is that the architecture is modality agnostic (it doesn't rely on language for its chains of thought), and I want to use it to bring COT reasoning to protein and anti-body generation. Basically, I hope for it to be the OpenAI o1 or DeepSeek R1 for domain-specialized scientific AI.
I’m building a quantum photonics experiment that is a variation of the quantum eraser.
One aspect that HN may find interesting is my use of Bayesian optimization to control and perfect key experimental settings. About a dozen of the wave plates and other optical components are motorized and under computer control.
Given a goal metric like "maximally entangle the photon pairs" the optimizer will run the experiment 50-100 times, tweaking the angles of various optics and collecting data. Ultimately it will learn to maximize the given cost function.
This sort of thing is commonly done with tools like Optuna during NN/LLM training to optimize hyper-parameters, but seems less common in physics especially quantum photonics. I'm using a great tool called M-loop to drive the optimization, which was originally developed for creating Bose-Einstein condensates.
https://github.com/michaelhush/M-LOOP
An iOS app made for saving locations. It’s called Pyxis: https://apps.apple.com/us/app/travel-journal-pyxis/id6639617...
I’m not a developer by trade but I’ve been learning iOS dev for about 6 years now. It’s become my project that I just keep working at since I personally use it a lot.
The app lets you save your favorite locations, add notes to them, add photos, check weather, tag them for better organization, and archive those tags for future trips. You can also mark off locations that you’ve been to already: think breweries or a coffee shop when visiting a new city.
For the next update, I’m working on a task list functionality for each location. The idea came as a shopping list based on which stores I go to but it can work for any other context as well. This way I can get rid of my shopping list from my task apps.
In terms of weather, I’m also adding historical averages to the forecast to have some sort of context to the weather.
Also leaning more into marketing these days (hence this post) and designing a new icon with some custom art work to give the product some sort of personality. I started learning affinity design to just do it myself so I learn some design software along the way.
Anyways, if you download it, I’d love to hear some feedback. :)
I'm working on https://www.hessra.net/
We've built a new auth platform with some new identity primitives and capability-style tokens using biscuits.
Right now, I'm trying to figure out ways to apply it and am looking into offering integrations with extremely fine-grained access control that wouldn't have it otherwise. So adding a fine-grained access layer in front of stuff like backend-for-frontend (BFF) systems, brownfield stuff with poor auth, or even OAuth stuff that just have really coarse scopes.
Are there any integrations out there that people want but the access control is bad for them? I'll build one for you!
https://replays.lol/clipper - A spin off from my main startup's product that lets Twitch streamers review their viewers' league of legend clips. Zero friction as we handle the recording of clips ourselves using bots that load up people's previous games using the League of Legends game client. Went live with my first medium-size streamer yesterday, excited to see how this goes!
Similar to Cameo but hyper specialised for league of legend streamers, if this shows some traction we'll expand to other games, and then to other industries (think a tennis star reviews one of your tennis points, beats out a generic happy birthday message?)
I'm working on an AI (Aethas, https://blog.aethas.ai) that preps context before my meetings, sets reminders, makes design drafts, summarizes meetings, and drafts follow-up emails without me asking. Basically trying to have an entire personal AI like Jarvis that has the same context as me but acts on it automatically. Uses my Obsidian for knowledge, Claude for reasoning, pull data from written notes... Still early days but it feels like magic when I use it.
I am working on Entangle https://entangle.cloud a way to add AI powered chatbot to a website.
I looking mainly for European market that is willing to self host your content or the model on your own servers, we will provide the technical support in doing so that and easily plugin the AI powered search or chatbot to the website.
But it is also possible to use servers provided by us to host the content and the LLM models.
Working on some new Bluesky tools and trying to learn more about the AT Protocol. Have been working on this following:
- Skyscraper, an iOS native app for Bluesky with focus on Liquid Glass UI. Launching hopefully in a ~week, and TestFlight available at https://testflight.apple.com/join/RRvk14ks
Then also working on a website/web tool that does the following: - A keyword/term notification service that observes the Bluesky Jetstream for usages of the term and sends email alerts.
- Provides an HTML/JSON backup archive of any Bluesky account. Quick way to archive popular accounts, politicians, public figures, etc.
- Trending Hashtag lists, to see what is trending the last hour, day, week, and month.
The web services all are available at: https://api.getskyscraper.com/tools/
Still working on my tool for creating custom Tailwind-style accessible color palettes for web and UI design:
https://www.inclusivecolors.com/
There's millions of tools that try to autogenerate colors for you using algorithms and AI, but they usually ignore WCAG accessible contrast requirements, don't let you customise the tints/shades, don't let you create a full palette of colors, and the colors often don't look right on actual designs.
This tool is meant to make customising tints/shades intuitive and quick enough in a few clicks via a hue/saturation/lightness curve editing interface that you won't want to rely on autogeneration. There's also a live mockup showing how your palette looks on a UI design that checks pairings pass contrast requirements, to guide you as you tweak your colors and to teach you the WCAG rules.
You can then export your palette to regular CSS variables, Tailwind, Figma or Adobe for use in your designs.
Really open to any feedback on features that would be useful! I think the only way I can make it simpler to use is to make it more opinionated about how your palette should be designed so interested in any thoughts about that too.
ChatGPT recommended this to me recently when I was trying to get some assistance with a usable Tailwind palette. I ended up not needing it right away but it's first in line next time I need to make one.
Any details about what problems you were having getting a usable Tailwind palette? There seems to be lots of different use case so I'd love to hear more.
That's awesome, I haven't kept up in what helps to get into AI recommendations. Guessing it's related to search result rankings? Not sure if the site would be in the training data. Curious if you asked about accessibility as that's my focus.
Gymlocity (https://gymlocity.com) - All-in-one gym management platform for small gym owners.
Building this for personal trainers with home gyms and small 1-2 location owner-operated facilities. The big players (Mindbody, etc.) are overkill and expensive for this market.
Core features: class scheduling, member booking, Stripe payments, and workout programming (the part most gym software ignores - trainers still use spreadsheets or generic apps).
Stack: React 19 + Vite frontend, ASP.NET Core 10 API, PostgreSQL, multi-tenant architecture so each gym gets their own branded experience.
Currently polishing the member dashboard and workout tracking UI. The goal is something a solo trainer can set up in an afternoon without needing to call sales or sit through demos.
I’m working on ApprovIQ, a tool for construction certifiers. Someone who wants to get something built will work with one of customers to get their project approved before construction. They’ll typically go back and forth on dozens of documents each of which must be checked over for compliance - drainage, water quality, fire safety, the list goes on.
Despite being intensely technical and detail-oriented, certifying construction is still mostly done by hand over email!
Our niche is full of folks whose lives we can improve with a portal for document management & comms, plus a sprinkle of AI for document understanding.
If you know someone - anyone - wrangling too many documents via email. Please reach out.
https://approviq.com
I’m working on Security Level 5 (SL5), which is basically “nation state grade security” for frontier AI systems. The core idea is that if a model’s weights or training artifacts could enable catastrophic harm, you should treat them like top-tier secrets and secure them accordingly.
One piece I helped with is SenL, a “sensitivity level” framework for AI labs. It’s like a practical clearance system for AI assets. Not everything in a lab is equally dangerous, so you label assets by sensitivity (weights, training data, eval sets, agent tooling, deployment configs, etc.), then tie that label to concrete controls like who can access it, where it can run, what logging is required, and what monitoring / two-person rules apply.
If anyone’s curious, SL5 is here: http://sl5.org/ and the SenL framework is part of the published artifacts.
> The core idea is that if a model’s weights or training artifacts could enable catastrophic harm, you should treat them like top-tier secrets and secure them accordingly.
My read here is that you're implying that if an attacker has access to, for example, weight data, they can invariably find a way to exploit it.
If that's a correct assumption, I think you're playing an unwinnable game, since attackers always have indirect access through inference of the model. It feels like locking down weights/training data/etc is the ai version of security through obfuscation.
Just my 2c, for what it's worth
Thanks for the insight!
I think this is exactly why some of the work is moving away from “assume unrestricted API inference forever.”
For example, we’re prototyping ideas like air-gapped or very low-bandwidth inference gateways, where interaction happens over narrow channels (serial, optical, audio, etc.), with explicit threat models and monitoring. The point isn’t that this’s practical for today’s models, but to reason about what inference might look like for AGI/ASI-level systems where the risk profile is fundamentally different.
Others are thinking along similar lines too. For example, this SPAR project on constrained and minimal inference pathways: https://sparai.org/projects/sp26/rec7NyTst8Upfp83l
I am devoting 2026 to focus on my rpg. The one differentiator here is something I am still not seeing: proper use of LLMs in games. The current batch is all lazy asset generation, maybe logic coding and what not, but not anything that could make the world actually feel alive.
Obviously, it all comes with its own sets of issues, but I am working through those as they come. But it is still a slow move solo.
Are you thinking more, of LLM based "AI" or "director" logic? Like an LLM tailoring the challenges and randomness the player experiences?
We did have some movement in the director logic before and, I forgot the zombie game in question, some of it seems pretty well served for the most part. I still think there is some room for a proper game master type that you refer to ( adjusting the challenge to what the player seems to be to do ).
I was personally thinking of making NPC less npcy ( not completely unlike dwarf fortress, but expanded ).
What is your 2c on some of the recent backlash about AI assets in games? Is that a vocal minority (game artists and similar ec), or is that part of a broader trend?
I think the artists have a valid bone to pick and I think some of the backlash is mostly coming from that quarter. The actual gamers are more repulsed by established studios using AI for asset generation, because it comes across as a lazy money grab. But gamers are both flexible ( will jump through a lot of hoops to do what they want to do ) and and set in their way of 'just want to have fun'.
On the other side of the spectrum are indies, who can afford to experiment a little more so you get more interesting uses ( like 'voicelines' by AI come to mind ).
I know what I want to believe, but it is hard for me to call it ( I think it is a temporary trend ), because I might be a little too close to it. After all, gamers have been conditioned to endure a lot over the past decade and mostly shrugged off most of the assaults on their favorite past time.
If I were to compare it to something.. it is almost like using LLMs for email summaries ( which is still what most bosses seem to be most pleased with ). There are better use cases. I think those were not explored yet.
https://sendbroadcast.net - I've been working on it since October 2024, and it's still going strong.
Currently on v1, but working on v2 release in the next month.
Still an AI time tracker that watches your screen.
Currently experimenting with a proactive agent, Don, that pops up like clippy and also works over email.
https://donethat.ai All the data security measures here: https://donethat.ai/data And other tools out there: https://donethat.ai/compare
Hmmmm. I am not sure I would feel comfortable running this.
I built an experimental system to test whether an autonomous AI can propose actions freely but be structurally prevented from executing side-effectful actions without explicit authorization. The system consistently blocks unsafe filesystem, shell, and network operations and produces a trace and diff proving nothing changed, even under adversarial pressure. The goal was to see if refusal and non-action can be enforced and verified at the execution layer rather than relying on prompts or logging.
CodjiFlo[1,2] - a code review tool inspired by Microsoft's CodeFlow, used by ~40,000 developers. It is especially tailored to power users of pull requests to improve contextual understanding and ease of code review and collaboration.
[1] https://github.com/pedropaulovc/codjiflo/
[2] https://codjiflo.vza.net
Over the holidays I built MakersHub.dev – a community platform for developers and creators building with AI tools. Whether you're a complete beginner exploring AI coding for the first time or an experienced professional showcasing advanced work, the goal is to have something useful: learning guides, project showcases, discussions, a tools directory, and news. Still early but actively building it out. https://makershub.dev
I’m interested in ocean container shipping, so I built a Google Sheets add-on to automate the process of tracking shipping containers. It’s called Container Tracker and I just got it live on the Google Marketplace. https://workspace.google.com/marketplace/app/container_track...
Main workload is done by the backend (serverless functions).
I am currently working on a HubSpot extension (that uses same backend) with the goal to target few other platforms where users work, and integrate the functionality into their daily workload, as opposed to having it as standalone website or mobile app. I have fun doing it.
mock, an API creation and testing utility. Any feedback is welcome!
https://dhuan.github.io/mock/latest/examples.html
I'm building a Postgres-native workflow engine using pgmq for queues and TypeScript worker. Workflows compile to definitions (SQL rows), letting Postgres orchestrate the DAG as state machines. The TypeScript DSL is type-safe with inferred inputs/outputs across dependencies with full autocomplete.
Declarative and functional in nature. Just a manifest wiring functions into a DAG and a Postgres SQL functions that manage the graph of state machines. Simple in principle and very opinionated.
Replaces 240 lines of manual pg_cron -> pgmq -> Supabase Edge Function boilerplate with 20 lines of explicit DAG definitions. Currently Supabase-only (leverages their primitives) but planning to make it agnostic for vanilla Postgres setups.
Live demo / explanation here: https://demo.pgflow.dev
As a former JPMorgan Chase engineer, rebuilding the Chase Travel UI so it doesn’t suck (there was never a role available on that team while I was there): https://rkdvis.com/chase-travel
I'm working on Increader, an incremental reading platform.
You put in all your bookmarks (also pdfs or epubs) and it puts them in a queue and tracks your progress. Read for as long as you want to and if you get bored with an article you just move on to the next one. Supports highlights and annotations as well as creating spaced repetition cards out of those annotations.
Really reduces the friction for me to start reading and it has made a noticeable difference to my media consumption throughout last year.
Started out as an exploration into the incremental reading concept, but it's become my primary interface for reading and I use it every day.
I haven't really talked about this to anyone yet, but it's getting to a point where it's polished enough for others to use.
It's currently completely free and you can try it without entering your email.
https://www.increader.com/
I've been building Nottawa for ~4y now and it's finally out in the open!
Nottawa's a free macOS app for making live audioreactive visuals. I'm trying to position it as a 100% free, batteries-included alternative to Resolume and TouchDesigner.
Not a tonne of users yet, but I'm hoping to get some traction in 2026. Would love love love to hear some feedback!
https://nottawa.app/ https://x.com/joe_crozier
Nothing as impressive as others here... but I have a few little things
--- Todo or else ---
The main thing is a new Go CLI called "todo or else". It scans your project for TODO comments and checks that they have deadlines
This allows you to hold yourself accountable for completing them
You can verify other things like whether TODOs have owners
I'm using Tree Sitter which means it can handle most programming languages (although, I'm shipping with a subset, as the grammars can get quite big)
I'm hoping to release this cross platform in a couple of weeks
--- Video: How PSX games were developed ---
A ten minute YouTube video about how PlayStation 1 games were developed
The tools, languages, and practices that were common in this era
Also the computers, software, programs used for creating assets (eg Irix Framethrower for SGI)
I'm hoping to produce a 10 minute video on my YT channel (jbreckmckye). Currently in the writing stage
--- Black Noise ---
A small, command line program for playing white noise
I think it would be a nice utility to have for focus sessions... maybe to double as a sort of pomodoro timer
Hey, I implemented a scrappy version of your TODO or else idea several years ago. Certainly not as sophisticated as what yours sounds like.
https://jezenthomas.com/2015/10/using-git-to-manage-todos/
I'm working on a "simple" (started out that way at least) ETL app for the public library system that I work for; the target output is SQLite databases where Datasette will be the platform to interact with the data being generated/extracted. Primarily, the goal is to provide a simple report tool for staff and give me a source for nightly snapshot outputs to a datalake in parquet format.
Continuing to build GhanaHousePlanner https://ghanahouseplanner.com/, currently focused on enhancing the cost accuracy and constraint solver for the generated house floorplans.
I’m building Shannot, a human-in-the-loop sandbox for AI agents on production systems.
Instead of filtering commands with heuristics (which agents work around), it dry-runs entire scripts in a PyPy sandbox, captures every command and file operation, then shows you exactly what will happen before anything executes.
I’ve just added checkpoint/rollback so you can undo changes if something goes wrong. Currently working on example scripts for common sysadmin tasks (nginx config, log cleanup, cert audits, etc.)
https://github.com/corv89/shannot
ScreenRecord.in: https://screenrecord.in I have worked remotely for most of my career and have found screen recording a useful tool to share ideas or ask questions. I liked the UX of Loom but always wondered if one could do it without installing any app. Turns out you can! Given today's browsers especially on desktops one can do a lot of things. ScreenRecord.in can record screen, webcam, mic and system audio if the browser supports them. Chrome has the best compatibility. All recording are stored in local storage I am figuring out what features to add.
I've got two going :)
https://shepherd.com/
Building a really unique book recommendation platform to help readers find books they might not otherwise know about. We just started working on a full book app that will be a private book diary, along with recommendations and insights based on your Book DNA. Think Goodreads, but rebuilt for readers who want a private space to track what they read, keep notes, and get truly personalized book recommendations (like Pandora or Spotify for books).
https://downforeveryoneorjustme.com/
This is a fun one I do with a good friend. Basically, to see if a website is down, or if it is just you, along with reported reasons from the community. We are working to add user accounts so you can create your own custom lists of websites/services to track
Updating my workout app with final touches. It’s local first, with self-hosted backup server as an option.
It’s text based, so it is basically an advanced editor/viewer for one long text note.
https://apps.yozy.net/swolog/web/
love the UX
I just completed a holiday project this week to measure the kids screen time using an Arduino device and a web app! I got frustrated trying to get so many parental control apps to play nicely, in the end this seemed like a bolder new approach to the problem.
I blogged about the whole build and coding project [here](https://partridge.works/screenie-christmas-project-2025-26/)
Code for the Arduino and also the web app is Open Sourced [here](https://screenie.org/get-device/selfbuild)
And you can play with it [here](https://www.screenie.org)
does it work with wife or just kids?
Continue to work on my project for remote private access: https://github.com/connet-dev/connet
Just released v0.12.0 which has a lot of package cleanup and some important bugfixes. Next, is making the relay infrastructure much more lighter, requiring less synchronization.
Personally, I'm using the hosted version[0] (which is just a repackage of the open source version with dynamic with tokens) to expose my NAS and syncthing web UIs to manage them while I'm away. Sometimes even through my phone (with termux)
[1] https://connet.dev
https://codose.ai
Still experimenting with different ways to make learning easier using LLMs.
I put together Codose as a tool where you paste a link to an Exercism or LeetCode problem, and it spins up a code editor with an AI tutor that walks you through the solution step by step, with mini lessons along the way when you need them.
You can try it without signing up but I’m on the Google AI Studio free tier right now, so I’m not sure how many uses can it handle
Building "human perspective maps" using non-linear dimensional reduction algorithms to visualize the "human value manifold" from hundreds of agree/disagree/pass statements, like Google Maps for complex large-scale conversations:
https://patcon.github.io/polislike-human-cartography-prototy...
Just paint with colors, click a painted group, and see what differentiates your painted groups. (Chrome on iOS has issues fwiw)
This is building on the philosophy of democracy-bolstering tools like Pol.is, which I've worked with (as a researcher/practitioner) for almost a decade
Trying to bootstrap a PKI certificate trust chain for facilitating trust projection and information verification online. Think of it as the ability to do something such as age verification at scale via a peer-2-peer ish mechanism instead of sending your government id to a porno service.
We are experimenting with Keybase key holders as CAs:
https://news.ycombinator.com/item?id=46576590
And also .gov email holders:
https://blog.certisfy.com/2025/12/using-gov-email-addresses-...
It's all self-service and requires no sign-up or download of anything, the app (https://certisfy.com/app) is an in-browser app and all the cryptography happens in the browser.
I am a passionate player of this traditional German card game called "Doppelkopf". It's a fantastic way to spend time away from screens. The game is quite complex which makes it super fun but also hard to pick up.
I'm building a small web app with an interactive tutorial and a browser-based singleplayer game that helps people learn and practice Doppelkopf. I've just released an English version:
https://doppelkopf.club/en
I am building an open-source, cross-platform desktop app for building and providing code-related context to chat based large language models. [1].
I found that sometimes I would rather interact with a chat interface to debug an issue or brainstorm architecture solutions in my repos. Agents are great for giving the model access to everything and letting it figure it out.
By manually prompting, it forces me to keep my metal model of the codebase up to date, and it allows me to provide just the context I want to the LLM.
[1]: https://github.com/haideralsh/prompt-lab
building out https://www.fluidposts.com, which is a writer for SEO folks.
Thought it would be a good way to learn about one form of marketing while building out some useful tools!
Looking for beta users and would love some early feedback!
Warenkorb+ (German for „shopping cart“): https://github.com/Lars147/warenkorb_plus
A browser extension (Chrome & Firefox) for simplifying my online grocery purchase workflow from Cookidoo to Knuspr.
I was tired of my weekly workflow of copying, pasting & sorting the grocery page for each item.
Also launched my first Hugo blog. Really nice experience so far. Wrote more detailed about the extension as my first blog entry: https://lars147.github.io/blog/
Developing a graph-based programming language called Graphoid. The original concept was to target AI-related programming. Probably not ready for that yet!
Repo: https://github.com/xvandervort/graphoid
Claude Code is doing the majority of the coding, with close supervision from me. I write notes while I'm working on it. Notes are here: https://www.patreon.com/cw/aiconfessions
Using Claude Opus 4.5 to query time-series data.
I have GBs of time-series data in a TimescaleDB database. It’s more complicated than this, but the gist is: I use natural language to ask questions about relationships in the data, Claude Opus 4.5 generates queries, and it finds patterns.
For example, I classify tens of thousands of news articles using different classification models. Then I ask Claude to write a query that tests for statistically significant changes in the time-series data at specific intervals after a given classification of article—and it finds patterns.
It passes train / test split validation. It will train on 2 years of data (2023 and 2024) and being able to effectively predict movements on the time series data using the classified news articles on the last year of data (2025).
We're doing so much planning and reviewing with coding agents like Claude Code and OpenCode.
I spent a day over break building a better UX for reviewing coding agent plans.
Plannotator - Annotate and review coding agent plans visually, share with your team, send feedback to the agents with one click.
Demo video: https://www.youtube.com/watch?v=a_AT7cEN_9I
https://github.com/backnotprop/plannotator
https://openrailwaymap.app
OpenRailwayMap is a project focused on displaying everything railway related in the world, powered by OpenStreetMap data.
I love it! I used it last month to find a track that recently got noisy in our neighborhood and learned a lot about the regional rail network.
PyTorch compiler and runtime for WebGPU https://github.com/jmaczan/torch-webgpu
Just released a Code Quality leaderboard based on the static analysis research I've been doing. https://codehealth.sibylline.dev/
Now working on comprehensive benchmarks for another tool I built, https://github.com/sibyllinesoft/scribe. Results thus far showing it reducing agent token usage by ~80% in real world tasks, but I need to repeat to get variance. Hopefully I can get a writeup out soon.
I've been working on the same tool since 2024 where I thought it might be a good time to build a tool for all the people who will build their own tools, eventually they will need to market it.
So I built a SEO/GEO Automation Tool for Small to Mid-Size Businesses who don't have a full-time team for that. [0]
The goal is to provide teams visibility across all the channels — Search and AI and give them the tools needed to outrank their competition. So far so good, the fully bootstrapped venture has grown over the last year and I've built quite a few big features — sophisticated audit system, AI Responses Monitoring, Crawler Analytics, Competitors Monitoring etc.
[0] https://seojuice.io
Adding a bit of context as well: This started out as a internal linking tool, but grew into something more based on the customer feedback — the database has now reached about 10TB of data about keywords, pages, AI responses etc, where I know who was ranking where and why.
And I'm trying to offer this "data advantage" to website owners, so they can grow, and also this is something that will be hard to replicate (at least quickly) with AI.
Eidetica, a planned decentralized database for local-first applications. I'm building out unstable features to get it to a point where I can show off the concepts but it's still fairly brittle.
I recently added better backend support for deployments, converted everything to async Rust, and setup Nix/Docker releases. I'm planning to build out some better example apps and workflows next, but everything will stay pre-alpha/unstable for now so that I can avoid getting locked in to any foundational issues. There are still a number of low-hanging breaking issues blocking the end-to-end usage which I'll need to address.
https://github.com/arcuru/eidetica
PlanMixPlay - https://www.planmixplay.com/
A live-performance software with a focus on creating 'musically connected visuals'. Currently, the biggest connectivity is probably tightly tied lyric visualizations. Some recent examples:
https://www.youtube.com/watch?v=mRHLzuUBz5o - She Wants Revenge - I don't want to fall in love
or if you prefer Mashups:
https://www.youtube.com/watch?v=e_Xq8Dh4NEw - The Lovemakers – Shake That 50 Cent (50 Cent vs. The Lovemakers)
https://interviewcop.com - interview cheating affects both businesses and legit candidates like myself.
I'm trying to solve this problem with AI agents that help interviewers to understand who actually can code and understand the code they're presenting.
2 projects around virtual tech events:
Open Tech Calendar, listing virtual tech events that include community participation: https://opentechcalendar.co.uk/
A listings site for virtually attending FOSDEM. The live streaming is great but the official site only lists sessions in the local Brussels time zone. You can choose your time zone here: https://virtuallyattend.teacaketech.scot/fosdem/2026/
I did a full circle: Graduated from doctoral studies, I'm working on automating science. Built an arxiv-like repo for science written by ai agents (https://ai-archive.io). To help scientists use this website and AI in their research, i wrapped opencode with ai-archive's mcp server and agents preconfigured. I then let people test this opencode bundle and contribute to the repo with a sandbox environment online (running opencode in container). Figured that authorative scientific repo requires grounding by real scientists and labs and therefore I am now negotiating implemeting automated science where I just finished my doctoral studies...
1. Zig-clr. "Borrow checking and other safety analyzer for zig"
https://github.com/ityonemo/clr
2. Molecular biology editing software. Will plug in to agentic ai workflows
I was working on an Arazzo[0] generator as a plugin for the Serverless Framework. This has now become a fully fledged runner. So far I have it using multiple OpenAPI source descriptions, and just about got retry rules working. Next steps will be for it to be able to reference Workflows in external Arazzo documents.
[0] https://www.openapis.org/arazzo-specification
Building a 22-service e-commerce platform in Rust (Runs on ~300MB RAM)
I'm 17. For the past 6 months, I've been diving deep into Rust to answer a question for myself: "Is Rust actually viable for complex, enterprise-grade backend architectures, or is it just hype?"
To test this, I built a full distributed e-commerce system (22 microservices, gRPC, Event-Driven) using Axum and Tokio. This is not meant as “how everyone should build”, but as an exploration of trade-offs.
Some hard lessons I learned along the way: Complexity vs. Performance: … The "Memory Shock": … (idle baseline; load benchmarks still in progress) Over-engineering: …
I'm currently squashing the final bugs…
I'd love to hear from senior folks here: …
Some time-constraints make it hard for me to commit to some open source work right now, but if I can I still improve my tool called "Software Manager". Goal for 2026 will be to have registered 4000 programs in total (I am at about 3750 right now) as well as make it dead-easy to interface with it through oldschool SQL (right now I just have .yml files and other flat files, though I can autogenerate a .sql database already as-is, but right now I don't query this dataset and it is not really optimised either). My long term goal is to make compilation on any platform dead simple. On Linux it works very well; on windows not quite as well yet.
Building CodeIQ - an AI tool that automates transaction coding for accountants and bookkeepers.
The interesting technical bit: it analyses your historic general ledger to reverse-engineer how you specifically categorise transactions. So instead of generic rules, it learns your firm's actual patterns - "oh, they always code Costa Coffee to Staff Welfare, not Refreshments" - that kind of thing.
Posts directly to Xero, QuickBooks, Sage, and Pandle. The VAT handling turned out to be surprisingly gnarly (UK tax rules are... something).
Been working on it about 6 months now. Still figuring out the right balance between automation confidence and "just flag this for human review".
Went down a DMARC rabbit hole last month or so. The holidays and some end of year commitments have kept me too busy to get much done, but I'm slowly tinkering with it.
I’m working on a context aware clipboard for Mac.
I found myself switching a lot between apps to get the same info, lots of copy/pasting.
Example, URLs in bookmark (which I forget about), project descriptions , images, folders.
So I built a Mac app that is similar to Raycast, but just for notes. If I want to save a webpage, I click control+option+C and then a window pops up to describe it.
If I press control+option+V, I get a spotlight like window where it does full-text search of all my notes and descriptions and filter so I can either:
- Open
- Insert the data into the current app (chrome, slack, ChatGPT).
I’ve been using it for a few weeks now, and not sure if others will find it useful.
I have been working on Print Relay[1], a cloud-based print relay system that enables remote printing to any printer connected to your local network.
PrintRelay consists of a cloud server and lightweight clients that connect printers to the cloud via WebSocket.
We use the excellent SaaS PrintNode at work, but about twice a year we have connectivity/routing issues between AWS ap-southeast-2 and their servers OS. PrintRelay is my attempt to not need PrintNode. Because of this PrintRelay is PrintNode API compatible.
1. https://github.com/schappim/print-relay
I'm getting ready to release a re-creation of an old mobile game that got delisted after the devs sold it to a company that subsequently ruined it.
The stack is Babylon + React + Capacitor, which was easy to step into as a full-stack dev with zero game building experience. Currently seeing what I can do to fix some performance issues, though it still works decently for a graphics heavy incremental/idle game.
Beta is still open for Test Flight. Can sign up via https://blossom-beta.blindsignals.io
meta-analysis, make Claude write a blog post, connecting and deriving set of concepts across a range of books, articles, papers, etc. Basically a syntopic reading machine. Its a meaty side project (and expensive), but I am curious how close I can push this. My current approach relies on a tool that I have built recently [0], it's an agentic memory but I am using a new memory model that is based on Zettelkasten principles.
What is really cool about it is that it natively capture connections between atomic ideas and evolve them. Which I believe it gets me one step closer to syntopic reading machine.
0: https://github.com/DiaaAj/a-mem-mcp
I built the running app I always wanted: https://runcoach.fly.dev
You get tailored running schedules and also some body weight strength workouts and healthy meals all in one!
Error Invalid input: 1 validation error for PlanRequest current_km Input should be greater than 0 [type=greater_than, input_value=0.0, input_type=float] For further information visit https://errors.pydantic.dev/2.5/v/greater_than
Yes it assumes a minimum running base, will improve thx
I've been working on binding raylib to s7 scheme and implementing a kanren on top of it basically just for the sake of understanding. I let AI write most of the binding code, though, because that is conceptually simple but very boring.
https://dawnofthe.dad/workout
Been working on a Google-sheet backed workout tracker, which basically makes it easy for me to see what I've done or not done recently and pick the next thing to do. I'm thinking of open sourcing this soon, but need to do some "de-monolithing" first.
I'm building https://swishfinance.app - minimalistic and simple stock widgets for your PC. That way I no longer have to make a conscious effort to check my portfolio — it's just there on my desktop.
https://gitswipe.com/
A mobile app for triaging GitHub notifications in seconds. Available for iOS and Android starting next week.
I 3D printed some new apertures for an infrared spectrometer at work to reduce some issues that cause artifacts in the data to enable higher accuracy measurements, particularly of high index and reflective samples. Seems to be working well now.
Very nice, did you run into any interesting challenges?
Mostly what I was able to read from the papers seemed to work. There were some remaining issues related to detector non-equivalence (I think) that appear to be solved with a temperature stabilized detector or very strongly attenuating the beam; the latter isn't ideal as it reduces the signal to noise.
It's interesting how fast such experiments usually run into the laws of physics. I'm trying something along those lines myself and I've found myself wishing daily that light traveled a bit slower. I see Discworld in a wholly different light ;)
1. Developing an voice AI agent (applied AI, not research)
2. Production ready AI - mix of code and human eval
3. Understanding and building for the new agentic AI commerce
Still working away on a free, completely offline transcription app that's available on all platforms with CUDA support as well.
https://blazingbanana.com/work/whistle
Currently tidying up some internal code (also removing the larger model on mobile platforms) and implementing proper diarization (who said what) so that it can be used for more than just personal dictation.
My iOS developer account is _finally_ approved so it will be available through the proper app store soon.
Just got started again on my "Poor Man's Bitemporal Data System" (discussed here: https://news.ycombinator.com/item?id=45118585 ).
Holidays nuked all the hot-cached context in my head. I spent a few days just spinning wheels until it repopulated. But the basic idea works now!
Much testing and benchmarking work remains to make sure it's not going to lose data, and that it won't denial-of-service itself (because object-map -> facts fan-out is big).
Also a second giant blog post is due (following the one discussed above). Lots of notes have accumulated.
It will be fun even if the concept ultimately crashes and burns to the ground :)
In which case, there's always datomic and xtdb :D
(edit: add note about upcoming blog)Love the idea! Read only the beginning of the first "giant blog post" so maybe it's discussed there, but in Sqlite doesn't WAL mode give you a kind of super simple, super basic temporal data system? If you somehow kept all WAL entries 'as is', it should be possible to reconstruct the state of the db at any point in the past?
Thanks!
> but in Sqlite doesn't WAL mode give you a kind of super simple, super basic temporal data system
Yes-ish.
For one, it is unitemporal. For another, it is for SQLites own transactions, not of the individual datums. Litestream is the way to replicate WAL into an object store / elsewhere. litestream-vfs is also looking good!
https://litestream.io/ , https://litestream.io/reference/vfs/ (announced here: https://fly.io/blog/litestream-vfs/ )
I'm trying to emulate the data assert/redact properties of append-only bitemporal-by-design data stores like XTDB. The giant blog post builds up the intuition from scratch. Or at least tries to.
So my system is going to be: bitemporal data, enforced by SQLite schema and application checks, along with WAL replication for point-in-time transaction backup/restore. Both are entirely distinct, and unaware of each other.
I’m working on a startup that aims to make decarbonisation profitable faster, based on the following idea:
Short and sufficient version here: https://doi.org/10.5281/zenodo.17288906
Extended version here: https://go.expinent.com/VlChn65
I'm working on interactive explorations of algorithms and machine learning. Small, visual, hands-on demos that help build intuition by letting people tweak code/parameters and see how things behave in real time.
Nice! Is it a website, an app, a Jupyter notebook,...?
I'm working on The Influencer AI(www.theinfluencer.ai), a platform that lets you create realistic and consistent AI personas('ai influencers') to use on social media or as your brand ambassador. You can use this unique AI person in photos and talking videos, for use cases like talking about your product, creating a high quality Instagram persona, photos for your e-commerce site, etc.
I'm adding drug-development features, and polish to my open-source molecule viewer and editor, Molchanica: https://www.athanorlab.com/molchanica.
Source code, in rust: https://github.com/David-OConnor/molchanica I've split out its building blocks into their own libraries on crates.io, for anyone building other bio or chem software. I don't think anyone uses them at this time.
Is Rust a common choice in this space?
No; I believe I am the only one doing this in the open. There are a handful of tools for bioinformatics, but am one of few building structural bio/chem ones.
For work related items I’ve been building out agent tooling for building some models and PoC projects related to energy industry applications. Been doing the consulting thing for a bit now and gaining more broad knowledge on some of these data center builds. Hoping to spin this into a product soon.
Started playing with gas town which is really cool. I had a naive version built that was just not good enough. This feels like a step in the right direction.
Haven’t had much time to work on any of my physical hands on hobbies lately but maybe when the weather gets better I’ll head back out to the shop again.
DTAP protocol.
Double TAP is lightweight testing framework where users write black box tests as rules checking output from tested "boxes". Boxes could be anything from http client, web server to messages in syslog. This universal approach allows to test anything with just dropping text rules describing system behavior in black box manner.
http://doubletap.sparrowhub.io/
I'm making a synthesizer drone generator (Webaudio API) with an audio visualizer (Three.js). It's a fibonacci sphere made of 100k particles. I send a sine wave down the spiral and it highlights the different spiral paths contained within the larger fibonacci spiral, which creates some really interesting patterns. I'm currently working on the audio component.
https://vimeo.com/1147473608?fl=pl&fe=sh
I am working on a query cost analyzer that looks at query performance and does maths against your infrastructure cost to understand what is your dollar cost per query. The idea is that teams (when tables are tagged appropriately) will be notified with their most costly queries and hopefully makes that actionable for teams to clean up some queries.
I am unsure if there is a need for a tool like this in the market but I am becoming more and more curious around databases so this felt like a lower barrier for my product-minded engineer skills to get into.
https://github.com/syamp/biscuit - Just a fun experimental tsdb completely written by codex.
Multiple things, but among many:
- A sports club management platform, and a way for end-users to sign up for sports events, lessons etc.: https://mojtim.ba/en/
- Given the raise of AI, I'm a hiking guide and would like to have that as an alternative, an outdoor activity agency - https://boa.ba (still very WIP)
I’m building a simple CSS redesign for Hacker News on Safari iOS as an extension: https://apps.apple.com/us/app/y-redesign-for-hacker-news/id6...
I have a few other small side projects that generally improve my day-to-day life, including a better calendar widget for shift workers and a video speed controller that floats on websites where I frequently watch videos for easy access.
A little desktop app that lets me upload transaction csvs from my bank and figure out how much I need to split with my partner. Mostly because I always forget to charge her for utilities or flight bookings and I hate going through the bank UI. Might also expand it with some simple subscription auditing logic.
Also, a dramatic anime intro (complete with cheesy AI generated theme song and video) starring our foster kittens. It's been interesting to learn about some of the techniques needed for consistency, how to storyboard, etc.
https://www.401k.live/
Public dataset for exploring 50,000+ 401(k) plans holding $7.5T in assets.
You can look inside a company like Google and see what employees invest in (mostly 2035-2055 target date funds) or how much they contribute ($30K - likely using Mega Backdoor Roth)
https://www.401k.live/plans/google-llc/
Still working on PocketWise (https://pocketwise.app), a simple double-entry accounting app. Just finished adding end-to-end encryption with a zero-trust server model. All encryption and decryption happens in the browser (using PRF), and the server only sees encrypted data.
Building some prototypes around the Recursive Language Model paper. I'm currently working on integrating it with Unity for scene and script automation. Thinking about standardized patterns for retrofitting this into other existing business systems. If I can make it drive Unity reasonably well, I think it could drive a lot of things.
https://arxiv.org/abs/2512.24601
My project is WithAudio and is gonna be WithAudio for a while. Its a text to speech reader. Initially I decided to generate pargraph by paragaph. But that was not a great call as users sometimes might have to wait for the whole paragraph to be ready before they can listen. Now I'm working on changing it to sentence by sentence. I think that + adding 2 new languages would take most of my January's budgeted time.
https://desktop.with.audio
Still working on my new LPFM radio station https://www.kpbj.fm
We are a 501c(3) and are actively fundraising to build a tower here in Shadow Hills and are launching our live stream and regular schedule February 1st. So far we have about 60 shows in the schedule.
If you're in Los Angeles and have an interest in radio, please hit me up.
A little static site generator using ruby templating engines:
https://github.com/LevitatingBusinessMan/edward
My Friend and I are working on Fostrom: https://fostrom.io
It's an IoT Cloud Platform built for developers. We're still in technical preview and are currently working on adding more telemetry to our small device agent written in Rust.
Check out our docs at: https://fostrom.io/docs/
Been hacking on https://rsolv.dev. It's a security scanner with a couple of unique twists; in addition to using AST validation to cut the false positive rate, it uses a heavily orchestrated LLM to write unit tests that fail if the detected vulnerability is present.
Happy for alpha users; it's really early days right now. Email in profile if you want to give it a try at no cost.
Writing a WebAssembly module parser from scratch, focusing on good diagnostics/errors and DX.
Currently it's fully-conformant to v2.0 of the spec, while I'm working towards implementing the recently released 3.0 version.
https://github.com/agis/wadec
Hello everyone, I'm new to this community. I was advised that this would be the ultimate place to have my Github repository audited?
I've live deployed an AI-constrained governance system on Ethereum mainnet - and I was hoping to share this with the community.
If this is NOT the right place.. please let me know if there is a better alternative in sharing my full repository.
https://github.com/thetaicore/tai-constitutional-architectur...
Two weekend projects:
- https://github.com/rcarmo/gotel (an OpenTelemetry tracing collector/UI, under heavy refactoring)
- https://github.com/rcarmo/toadbox (a simple Docker-based agent sandbox to run Toad/OpenCode/Mistral inside, which I've been cleaning up for general use)
Locque is a language designed to both be easy for LLMs to read and write, but also for humans to review, and it's designed along my philosophy that the more structure you can give coding agents, the better, so it's dependently typed and hosted on Haskell at the moment. Feedback eagerly solicited.
https://github.com/jaggederest/locque
TCKR - Scrolling Stock LED Ticker for Windows https://github.com/krypdoh/TCKR
I've been working on this for several weeks/months and I'm happy with the result!
Vibe coded it as my programming skills have eroded with time (or they never existed).
I would really appreciate some feedback.
A growth marketing Agent that's connected to your Google Analytics, Search Console, and can crawl the web to analyse your competitors: https://refreshagent.com
https://inSolitaire.com
I'm enjoying building a website with solitaire and puzzle games.
I am currently rewriting the engine for the fourth time and plan to add 400 games to the platform in the coming months, as well as social features such as daily challenges, awards and leaderboards.
My main goal, however, is to make this project the largest collection of free modern solitaire games available for mobile devices and desktop computers.
So far, the project has been incredibly exciting, and I've learned so much!
ALT (https://www.APILoadTest.com): I recently built ALT as a no-code platform to load/stress/performance test APIs with pre-defined (and customizable) load profiles, via globally distributed machines. The goal is simple: key performance metrics, in seconds, without scripting. I also added the feature to directly import collections from Postman.
Here's a walkthrough of how ALT works: https://youtu.be/m2MER0KW3yA
https://bttrne.ws/
HackerNews with a better UI (same content)
I’m working on PolyMCP a simple and efficient way to interact with MCP servers using custom agents
I’m a remote dev building an iOS app called Springus. You add a few pics / your closet basics, and it suggests outfits based on what you already own. It’s less “fashion app” and more “tiny morning momentum hack.”
Would love feedback from other WFH folks — a weird amount of my productivity comes down to how I start the day.
I am working on a camera module that has SLAM built-in: https://x.com/_asadmemon/status/1989417143398797424
Running on a single-core armv7. It includes a VIO and a nice loop closure. I am now optimizing it further to see if I can fit some basic mapping too.
I am working on version 6 of the Nutrisense. A device to keep track of what you eat (calories/protein, etc.) without the need to type it in, in some app.
https://marcelv-net.translate.goog/index.php?w=apparaat&id=3...
https://ardent.pet/
My Family recently (in the last couple of years) started to breed Ragdoll cats in the U.K.
In an attempt to support what's involved in this I built Ardent for them. It covers a bunch of the day-to-day concerns (weighing and health tracking), Lineage and Inbreeding prevention, and Owner Pack generation for handovers to new Owners.
I've been experimenting with what a serverless cloud platform would look like that networks wasm components using gRPC:
https://github.com/vimana-cloud/vimana
A Substack for 80/20 life advice that gets repeated every year. We all know what to do, the idea of this is to remind us every week of one aspect of life and what best practices are for that area.
https://euzoia.substack.com The concept: https://euzoia.org
Tried to do this as low tech as possible, so website is just an off the shelf notion wrapper
Platform for apps, currently Lua support ( future WASM). Think like apk (android app) but spk.zip that has server.lua file for backend and htmt/js/css that get served for front end plus platform gives Files, SQLite db (namespaced/scopped access ), Auth, User management All in single static binary Cz go
Edit: formatting
Lost link in editing
https://github.com/blue-monads/potatoverse
I've been experimenting (over 130 test batches so far) with creating an all-purpose balm (lip, hands... stuff).
https://workingbalm.com
I wanted something more useful to carry than typical gooey lips balm full of petroleum and silicone.
I have been working on the next generation of Canva and Photoshop for highly regulated verticals where there are specific demands placed on the generation and edit flow.
https://hawtads.com
If you are a brand who needs to deploy advertisements at scale, don't hesitate to reach out.
Building Cassette: a dead-simple computer in the cloud: https://cassette.sh
One always-on Linux box to run apps, databases, CI, and AI agents without hyperscale complexity, surprise bills, or Kubernetes. AI-driven app explosion plus mature open-source deploy tools make simple servers fast, cheap, and fun again.
I'm working on Narwhal, a pub/sub protocol and server specifically designed for edge applications. The protocol can be extended via an external component, and the product is completely free and open source: https://github.com/narwhal-io/narwhal
The easiest and simplest calorie counter
https://www.journable.com/
I am working on an failure proof introduction to Lisp/Scheme for imperative programmers. I always thought there was something unapproachable about Lisp. But I finally figured it out, and I want to document my path for people like me. It's going to be great. Expect to be done in a week or so!
I am super curious about tensor parallelism and the mechanisms behind how some models can activate only some of their attentions.
Made a landing page: https://shipmentplanner.com
Now I'll have to bite the bullet and start working on marketing!!!!
I'm having fun playing around with AI by building a coding agent with both a CLI mode and a web UI https://github.com/martydill/flexorama.
Building something is a nice break from the corporate world.
A personal agentic tool for action cam video editing. When I moto / ski / bike, I leave my helmet cam rolling the entire time. It takes a very long time to 1. find scenes and 2. edit them down into a reel or video. It works well for my amateur use case - basic edits with only cuts and transitions.
I’m working on agent-os: a slf-hosted web UI for managing multiple AI coding CLI sessions (Claude Code, Codex, OpenCode) with multi-pane terminals. All free and open source: https://github.com/saadnvd1/agent-os
I am working on SIP4AI [1] (a VOIP soft-phone where AI is on the other end, not a human). You can self-host it, register a SIP line, and let AI take the calls.
It works with your existing phone system, so you can just add AI as a line without having to replace everything…
1. https://sip4ai.com
A new numbers game named Sigma. It’s like a crossword plus Sudoku (kind of). Try it out? https://jm.cool/games/sigma
Finally managed to find some time to implement encryption for envelope [0]
[0]: https://github.com/mattrighetti/envelope/pull/46
I wanted to learn a new(ish) skill and am building a button box for sim racing with Arduino Nano, momentary switches, etc. It has been about twenty years since I’ve touched a soldering iron or built any kind of breadboard circuit.
I’m a solo entrepreneur - working on a questionnaire solving tool (think security questionnaires, RFPs and similar). Would love to hear any feedback.
https://requestfx.com
I threw together a simple patch-style synth to play with the Web MIDI APIs and React Flow. It’s an early experiment and part of my personal “try things and learn” site.
https://www.futile.com/patch
An Android app for CORE (YC S23) to make it as easy as possible to capture data to Core on-the-go. Voice-based UI, also quickly type in memories, and later retrieve by asking. https://memory2.app
I am working on invoicing web app https://www.haiku.lt . Currently focusing on marketing and EU e-invoicing part.
I'm still working on my PDF search engine for desktop: https://www.docgoblin.com/ I'm implementing a bookmark utility right now and hope to add support multiple E-books format in the near future.
A one time payment app - interesting (I'm also working on something with similar moneytization solution). How are things going? I'd love to know the experience of another solopreneur, what stack are you using? I wonder - What are you using to parse PDFs and extract the text? I found that is a nightmare when was doing something similar for WithAudio (my app). - Are you just extracting the text or you are doing any post processing to identify which lines belong to the same paragraph or not?
Things are going slow, but it is a passion project so it's ok :) A few people have bought a licence and it seems most people who try the app are very happy with it so I'm happy too.
The app is entirely in Java, with javaFX for the UI and Lucene for the search engine. To read and render PDFs I use PDFium.
Turtle graphics with lambda functions. I don't know if it has any interesting applications, but I wanted to learn lambda calculus
Thanks to Qwen VL, I shared my grandmother’s recipes on the web.
https://lescahiersdejeanine.fr
I am trying to get the TLA+ tools to run completely in the browser https://github.com/tlaplus/tlaplus/tree/master/tlatools
https://www.querybear.com - basically retool but make creating tools SUPPPPER easy
Publishing updated FreeBSD AMIs every time there is a security or errata update. This will make life easier for FreeBSD/EC2 users since they won't need to wait for patches to download when they first launch an instance.
Hacking together a little 3d snake game for my steamdeck, heavily “inspired” by that spherical snake post last week.
Never published to Steam before, it’s been a fun learning process.
A very simple facebook-ish social media PWA that is meant to be easily selfhosted on low end raspberry pi level hardware, bun + sqlite + mantine.
My friendgroup has gotten increasingly concerned with the gradual enshittification of various services we depend upon, and are looking at various alternatives. In some cases there are good selfhostable options (nextcloud, mattermost/zulip), but I decided to write my own tiny PWA to cover facebook-like needs.
The goal isn't really to scale to >1000 users, just to be simple to spin up for a small group and be easy to manage. I'm hoping to run multiple instances, eg one for family, one for college friends, one for local friends, etc.
My process has been pretty ADHD though. I recently read the phrase "It doesn't have to be done, it just has to be perfect" and felt personally attacked.
I've just released sanctum 1.0.0 and am taking 2-3 weeks off from hacking on it while I work on a little dumb game to be creative in a different way.
https://greatcountry.org
Comparing world countries on as many uncorrelated statisticals factors I could get my hands on.
Fun fact: in overall top 10, there is only one country that is not in Europe.
I'm working on an app to make testing available to all brick and mortar retailers (proofpod.ai).
The most difficult technical challenge has been designing a pipeline to fully automate choosing test & control locations using synthetic difference-in-differences.
Making rent as an open source developer.
Attracting new monthly sponsors and people willing to buy me the occasional pizza with my crappy HTML skills.
https://brynet.ca/wallofpizza.html
Hi I have built ThreadsDelete.co
A tool for bulk mass delete posts, replies, Quotes and Reposts for Threads platform.
Still adding new features
Echoresponse - a tool for responsible disclosure. Security Researchers and companies encode some of their secret knowledge in LLMs and the LLMs have a discussion and can say one word from agreed upon list back to the party that programmed them.
- https://dict.press - A (language) dictionary publishing and management system.
- WIP: A FOSS, self-hosted Luma alternative (for use across our community initiatives)
Rate my professor for my university because it didn't have one: https://www.ratedeeznust.com/
building a rest APIs that aggregates and structures data from the www
- ETL is vanilla Python - Orchestrated with Cron and SIGUSR1 - http is Nginx -> uvicorn-> FastAPI
Data lives in a CAFS indexed with Xapian and my current time to load is ~200ms
It's been A LOT of work all far. I came in with some skills and have learnt a lot along the way
At the moment I’m building a succinct data structure library, doing one algorithm at a time. There have been some very impressive papers came out recently. The numbers look promising.
I am working on a new Unix-style filesystem for the browser, in https://exaequos.com For local and remote storage and encrypted
Working on https://greatreads.dev/ A place to aggregate and find articles from developers' blogs.
Always on the look for new sources to be added. If you have a blog or read any that you want to recommend. Just let me know.
If I am missing something, isn't this the whole purpose of hacker news to be a link aggregator for technical/developers?
Hacker News is mostly anything tech related. That are links from news websites, github projects and etc.
My project is just for reading individual blog posts from fellow developers. There is also a lot of filtering capabilities as well.
I’ve been working on a fontdrawing app that only uses the keyboard (inspired by vim) for a couple of months. Finally got export to ttf working this weekend and got to see a bad looking j I drew in Figma
tirreno ~ open safety platform
Github: https://github.com/tirrenotechnologies/tirreno
Live Demo: https://play.tirreno.com (admin/tirreno)
Docs: https://github.com/tirrenotechnologies/DEVELOPMENT.md
Little Snitch clone for Windows and Android.
Ooh, do share when ready
Thinking of new ways to visualize market data in 3D like this helical candlestick chart:
https://cybernetic.dev/helix
Love the matrix UI with the minimap!
Thank you, I appreciate it! I'm also planning to add a graph visualization of the correlation matrix.
If you work in machine learning, we are building a much more powerful and faster Neptune/Wandb replacement.
https://minfx.ai/
Can it be hosted locally? Would be great to beat mlflow
We are planning to make an offline local version, see also https://minfx.ai/why.html and Web + Native
For personal projects:
Used ai to create my own mind mapping tool for private use.
I also created a private cursor-like / loveable-like tool that I can use for my own vibe code prototyping on the go with my phone.
Still working on https://cartes.app, the libre Web european Google Maps alternative.
https://battlecode.org/ is in January, I look forward to it every year.
This is awesome! I can't believe I've not heard of it before. Thanks for sharing!
working on a free mini-game for f1 enthusiasts, completely build by claude sonnet 4 and hosted on AWS
consistently getting traffic of 30k in a month with peaks upto 10k in a day
This has been on for a few months now but I'm thinking to add new features as users are asking for multiplayer support. Would love some feedback
[0]: https://www.hotlapdaily.com
I am currently working on developing a tool for analyzing AWS Costs, It's currently in progress I plan to share soon
Making complex simple again.
https://unrav.io
Made a simple anonymous feedback app - https://feedbackok.com/
A Python terminal audio recorder. Using pyaudio.
gthumb.ai
AI models and open source robotics for food production.
From backyard gardening, subsistence farming, urban gardening, and other forms of small scale agriculture.
We believe no one owns nature and that all growers have 100% right to repair any equipment we offer.
Our first IOT device (greenbox) is in an open beta for 2026. Please reach out to support@gthumb.ai if interested.
testeranto.com
I have built a multi language bdd test framework. A human can write the bdd specs and an LLM will generate the code to match.
I'm working on a negotiation tool. A sidekick for negotiations. No public domain yet.
Building a text editor w/ git-like version history
versionary.app
Not really a side project but:
I've been using linux for a few years now as my main/only OS, but have mainly just used Linux Mint as a sorta plug-and-play distro.
Looking to revive my 15 year old ThinkPad (1st laptop ever!) by building up from a base Void linux install. As I'm doing it I'm writing install-scripts and getting my dotfiles in order (after never really doing so for 17+ years as a programmer lol), so I can repeat the process in the future on other machines, or when I want to do a fresh re-install.
https://github.com/staydecent/void-setup
Also building a CW Ham radio from a kit
I am working on vulnetic.ai, an agentic penetration testing platform.
Finally writing up the documentation (architecture and safety concept) for the fly-by-wire system for the homebuilt airplane I'm 15+ years into designing. Got to OML lock about a year ago, and the aerodynamics are checking out, so really hoping that I can get a subscale flying in 2026 (although I've said that before). On full scale, major remaining design task is structures, but there's plenty of other stuff (propulsion integration) as well.
Repo-manager, SBOM tool for managing dependency (like Artifactory)
launched https://www.cofounder-hunt.com today
Finding other co-founders based on proof-of-work
A Gradle plugin that makes developing Java modules-based libraries super easy: https://news.ycombinator.com/item?id=46572154
We're at Open Community organizing an AI Vibe Coding Hackathon. As of now we got near to 500 builders, big Sponsors like ElevenLabs, Daytona, Nord Security brands and etc. Near to $200,000 in Prizes. Everyone is welcome to the hackathon https://vibe.devpost.com, Technical or Non-technical. Since it's vibe coding hackathon. Our discord btw - https://discord.gg/nUdcd9p8Ae
querybear.com - basically retool but make the learning curve waaaaay lower
Open-source SaaS for every vertical
Every vertical?
Yes, we just launched our Shopify alternative. Next is Toast alternative for restaurants, MindBody alternative for gyms, etc. We’re going to leverage these platforms to build a decentralized marketplace around those verticals.
I've been thinking about building a compiler for a while, but didn't have much time to do it. Over the break, I finally built a significant portion of a working, optimizing compiler. Heavily built by AI, of course.
https://blog.paulbiggar.com/full-optimizing-compiler-with-ai...
Futzing around with analog televisions.
I relaunched one of my Dutch agricultural communities to reach a more international audience. I’m starting to see great traction and it’s very rewarding: https://www.tractorfan.us
Doing Math Academy
IQ Test Platform (https://www.riotiq.com) I've been working 3 years on this. We developed our own professional, modern, proprietary IQ test and IQ testing platform for individuals to test themselves and to administer professional IQ tests to others. Lots of silicon valley folks seem to love it for their startups and hiring, but we built it mainly for IQ researchers from the International Society For Intelligence Research (ISIR) and psychologists.
Puzzleship - a free daily puzzles website with the archives paywalled. Right now it has Logic Grid Puzzles, Zebra Puzzles and a dozen of Solitaire variations. I'm pretty proud of the LGP generator algorithm and some experienced players also liked the way the puzzles are constructed. Launched in early December last year.
https://www.puzzleship.com/
An RSS reader / dashboard.
It has AI summarize buttons (gemini-flash-lite is so fast!) along with other features I wanted. I'm almost done adding a "war mode". The user (me!) specifies a list of OSINT style x users which show up sequentially in a grid along with a ticker on the bottom of polymarket markets I've chosen. War mode is also obviously only available in dark mode...
A lot of things because LLMs enable my potential ADHD and no-filter-brain. A thing that controls a small FPV drone from the PC with vision input. Cloudflare but in Europe! An open source synth out of a Pi Pico! A reservations engine for local market. A small orchestrator for when Swarm is annoying but K3S is just too much!
I'm finally revisiting a distributed syscall model for transparently scaling unix instances. syscalls get translated into batched operations on an underlying non-transactional datastore. On the service side, database operations get backed by a proxy serving whatever filesystem or socket interface you like. Scaling is one motivation, but the ability to enforce fine-grained policy on these data operations is another big one.
MOOLLM -- treating the LLM as eval() for a microworld operating system.
Started this incarnation on Dec 30, 2025 -- but it's the crystallization of decades of earlier prototypes, all the way back to my Commodore-64 Logo Adventure. Built on top of Anthropic's Skills framework, extended with seven innovations (and counting):
1. Instantiation -- Skills as prototypes creating instances with their own state
2. K-lines -- Names as semantic activation vectors (Minsky's Society of Mind)
3. Empathic Templates -- Smart generation based on semantic understanding, not string substitution
4. Three-Tier Persistence -- Platform (ephemeral) → Narrative (append) → State (edit)
5. Speed of Light -- Many turns in one call, minimal tokenization overhead
6. CARD.yml -- Machine-readable skill interfaces with advertisements
7. Ethical Framing -- Room-based inheritance of performance context
Lineage: Colossal Cave → TinyMUD → LambdaMOO (filesystem as world). Papert's Logo and constructionism (learnable microworlds). Will Wright's SimCity and The Sims (I worked on the originals) -- the "Simulator Effect" where players imagine more than you simulate, and the SimAntics visual behavior programming language.
YAML Jazz: Comments aren't ignored -- they're semantic. The LLM reads and interprets them. A comment like "# gentle but firm" on a character trait actually affects behavior. This inverts the traditional "comments are for humans" assumption. Comments become part of the program and data.
The core idea: instead of prompt engineering, you give the LLM a github repo filesystem to inhabit: a persistent microworld. Seymour Papert's Constructionist philosophy comes alive, with Minsky's K-Lines pulling the strings. Skills are programs (not documentation). Characters have persistent state in directories, and can reflect on and edit themselves. Everything is inspectable and editable by human AND model. Model and platform independent. Runs on Cursor and other tools and orchestrators.
The proof is in adventure-4 -- a complete text adventure with 150+ files, 6000+ lines of session transcripts.
Repo: https://github.com/SimHacker/moollm
MOOLLM Manifesto: https://github.com/SimHacker/moollm/blob/main/designs/MOOLLM...
The MOOLLM Eval Incarnate Framework: https://github.com/SimHacker/moollm/blob/main/designs/MOOLLM...
Adventure 4 Example: https://github.com/SimHacker/moollm/tree/main/examples/adven...
My sessions as proof it works: https://github.com/SimHacker/moollm/tree/main/examples/adven...
79 Anthropic Skills (standards compatible, plus extensions, intertwingled with k-lines) and growing: https://github.com/SimHacker/moollm/tree/main/skills
A guided tour through the MOOLLM skills and microworld -- Session Log: K-Line Connections Safari: https://github.com/SimHacker/moollm/blob/main/examples/adven...
Adventure Compiler Design Discussion -- Adventure Uplift Session Log: https://github.com/SimHacker/moollm/blob/main/examples/adven...
MOOLLM Kernel: https://github.com/SimHacker/moollm/tree/main/kernel
Happy to answer questions about any of the weird design decisions!
Is anyone still working on functional immortality? Are we going to get SAGI before it? How is the head/body freezing scene nowadays?
EV recharging software. book on HPC. book collection of short stories. Golang latency instrum lib. realtime Rogue-like game set in postapoc NorAmerika.
Just about done building a C# wrapper for libcurl targeting Windows 98. Then i can finish my system tray weather app for Windows, which will work on all versions of Windows from 98 to 11.
Another camera body around the RPi HQ cam and updated camera software.
I've been buying vintage lenses to try out.
A bunch of things that get zero engagement whenever I mention them in one of these threads
https://soatok.blog/2025/10/15/the-dreamseekers-vision-of-to...
I'm a filmmaker, and this is ArtCraft:
https://github.com/storytold/artcraft
AI tools are becoming incredibly useful for our industry, but "prompting" without visual control sucks. In the fullness of time, we're going to have WYSIWYG touch controls for every aspect of an image or scene. The ability to mold people and locations like clay, rotate and morph them in 3D, and create literally anything we can imagine.
Here are a bunch of short films we've made with the tool:
- https://www.youtube.com/watch?v=tAAiiKteM-U (Robot Chicken inspired Superman parody)
- https://www.youtube.com/watch?v=oqoCWdOwr2U (JoJo inspired Grinch parody)
- https://www.youtube.com/watch?v=Tii9uF0nAx4 (live action rotoscoped short)
- https://www.youtube.com/watch?v=tj-dJvGVb-w (lots of roto/comp VFX work)
- https://www.youtube.com/watch?v=v_2We_QQfPg (EbSynth sketch about The Predator)
- https://www.youtube.com/watch?v=_FkKf7sECk4 (a lot of rotoscoping, the tools are better now)
cool stuff!
Thank you!
If you give it a try, I'd love to get your feedback. I'd also like to see what you're making!
Navigation.