I'm working hard to prepare for my side hustle, hoping that one day I can support myself with it. Screw that shitty boss and working overtime every day.
Currently, I'm iterating on two websites:
https://aivideopromptgenerator.com/enhttps://www.fast3dtext.com/
Take it slow.
I am developing an augmentative and alternative communication (AAC) iOS app for non-verbal adults that uses LLM-generated phrases to speed up communication. My father had throat cancer and lost the ability to speak. He was prescribed a standard grid-based AAC app but couldn't build the motor-memory needed to navigate it quickly. (https://www.speak4me.ai/)
I used to use squeaky.ai which was a pretty simple session recording tool, but they got tired of running it and shut it down. PostHog is pretty great and with a generous free tier, but I get overwhelmed by their huge UI since it's such a large product now.
The core of the product is session recording, which I implemented using the rrweb library, which itself uses the Mutation Observer API. Everything else is pretty standard React/TypeScript/TanQuery for the front, Java/SpringBoot/PostgreSQL for the back.
No users, but working through the journey of trying to become a better copy writer / product manager / marketer! I've been getting some great advice on how to improve my landing page which has been exciting.
Have you checked Microsoft Clarifty - https://clarity.microsoft.com - it allows to generate heatmap and do session recording as well. And it is free. What are advantages of https://scryspell.com over it ?
I'm working on Habitat. It's a free and open source, self-hosted platform for communities to discover and discuss their local area. The plan is for it to be federated.
I'm building Concert Map [1], an interactive map of live music events all over the world.
I was frustrated with how difficult it was to find concerts and festivals when travel plans are flexible. Questions like:
- Is there any show I could catch on my next cross country road trip?
- I want an excuse to visit Seattle this summer, is there a good weekend I could pick where I could see live music there or along the way?
- I like to travel for one big music festival per year, is there a good match for me anywhere in the country, or abroad?
The idea behind concert map is to show all events on a map, and highlight the "most relevant" events for any area, so you can skim through the map and find events that are worth the travel. The "relevant" events depend on the view context. If you import your top artists, then the most relevant events are ones with artists that you listen to most. Otherwise, it's based on global artist popularity.
I'm currently working on integrating with more event providers, and working on loading performance. I've been using leaflet.js [2] which has worked great, but I'm experimenting with switching to supercluster [3].
I built it after a conversation with a friend thats a RE agent.
As an agent, he was literally spending 45+ minutes per listing just staring at a blank screen trying to make "3BR/2BA ranch" sound compelling.
The AI actually analyzes the photos to pull out key features he might miss and adds neighborhood context, like walkability score and gets professional descriptions in 30 seconds.
Just launched it as a standalone SaaS - would love feedback from anyone in the RE industry.
I've gotten a lot of use out of Claude Code in the past few months. The big problem for me is struggling to stay on top of multiple sessions in parallel. I know that all the cool kids on twitter have all these AI employees and agents farms... A lot of its probably bullshit, but directionally its clear that tooling will absorb some fraction of my workload.
I know I need to push myself to have 2 or 3 balls in the air. But without push notifications to know when Claude Code needs permissions or input, I end up "wasting" time a lot.
So having an app that sends discrete push notifications to my watch is pretty nice.
You talk to a voice agent, then voice agent talks to Claude Code. The voice agent acts like a buffer for me to ramble and gather my thoughts. Then the voice agent writes the prompt for claude code when I'm ready. So that unlocked walk and code for me. Now as I walk to the train from the office, I can dump some of the brain ideas into Claude Code plan mode, and they'll be waiting for me in the morning.
I'm trying to change the nature of LLM compute, with an idea that's been scratching at the back of my mind for decades. If it works, someone (likely not me) will make Billions, or even Trillions of dollars. I'm very deep in eccentric old man/lone genius territory. Current odds favor the former.
BitGrid[1] - I'm trying to figure out how to do software for a BitGrid. Long, long ago I looked at computing, and decided that the best high performance strategy was something like an FPGA, but simplified, and optimized around overall performance, and not simply the lowest latency and most efficient use of silicon towards that goal. A vast 2d grid of 4x16 bit look up tables, with latches, seems to be optimal, to me. (It's a hunch, that I could very well be wrong about)
Getting an expression translated to a directed acyclic graph of binary logical operations seems to be something that's been achieved, with extensive help from ChatGPT5.[2] (Warning, it's a huge pile of stuff almost exclusively written by an LLM, not me, as I say at the start of the README.md)
The thing now is to figure out how to map that graph into a grid. The LLM has certain strong opinions about how things should go, using the phrase "physicalize" that have taken it off into a few rabbit holes I've gotten us back out of. I think I'll get us there, but not as soon as I had hoped.
The latest is that just getting bits to move across a grid took some backtracking, but forward progress is being made again. It seemed to think that you can just plop data into the middle of a grid (which required defeating the one advantage of a bitgrid... oops)
The thing I really want to know, is how big a BitGrid would it actually take to hold a frontier model LLM like ChatGPT5. Obviously it's not going to fit on an single chip, it's going to be distributed. I've actually managed to figure the distribution and communications part out. It's just the raw size and a good estimates for the current power requirements, and area of a cell implemented in 7nm or better silicon, and the math falls out the other side. You either save billions of dollars, and gigawatts, or you don't.
How? Imagine a grid of FPGAs big enough to hold ChatGPT5, with all the math programed directly into the logic, not separate ram/matmuls like a sane person would do. Add latching periodically through the flow so that it's effectively a 4,000,000 stage pipeline or so. If you then run it at 1 Ghz, you'd have a maximum throughput of 250 tokens/second. However, you could then run 3,999,999,999 other streams with the other slots through the pipeline. ;-)
I was hoping to get the above software figured out before the submission deadline for the TinyTapeout[3] Sky25a run in 18 days, to get an ASIC made. I think I might have to start working on that submission before I even know if it's worth it. My hunch, and every bit of math I've done before says it will be, but it would be nice to know for sure.
I hope to have some estimates figured out by next month, and will let you know if I made it into Sky25a or not.
One of these again.. love to see all the projects, but man time flies when you're not doing much
Still slowly expanding https://wheretodrink.beer with new mini-features I want (like the obscurely hidden https://wheretodrink.beer/near-me - it probably need an interactive map), and adding on new venues. Struggling with getting any sort of organic traffic, but not sure that's even something I want..
Been toying with creating some other projects; one for helping with lost+found for SMBs in hospitality, and another with administrating "co-operatives" (collecting dues, coms, automating legal requirements).. So naturally you start building all the scaffolding that you could have bought, like gdpr-compliant privacy first multi tenant authentication.. That's fun, but I should know better.
https://codeinput.com - a tool to augment Git-related processes (merge conflicts, ownership, automation, etc..) I've been working on it for close to 10 months and it's still not to launch. On the other hand, I've been building it fully on Cloudflare Workers (WASM/Rust) and the whole back-end is a single wasm worker.
Continue working on https://AllZonefiles.io. New features in August 2025:
- daily download only new domains: https://allzonefiles.io/new-domain-lists
- daily download only expired domains: https://allzonefiles.io/expired-domain-lists
- download all latest domain lists as one ZIP file: https://allzonefiles.io/all-zones-zip
Not much, but these features were requested by users a lot.
I'm working hard to prepare for my side hustle, hoping that one day I can support myself with it. Screw that shitty boss and working overtime every day. Currently, I'm iterating on two websites: https://aivideopromptgenerator.com/en https://www.fast3dtext.com/ Take it slow.
I am developing an augmentative and alternative communication (AAC) iOS app for non-verbal adults that uses LLM-generated phrases to speed up communication. My father had throat cancer and lost the ability to speak. He was prescribed a standard grid-based AAC app but couldn't build the motor-memory needed to navigate it quickly. (https://www.speak4me.ai/)
https://scryspell.com - a simple session recording and analytics tool.
I used to use squeaky.ai which was a pretty simple session recording tool, but they got tired of running it and shut it down. PostHog is pretty great and with a generous free tier, but I get overwhelmed by their huge UI since it's such a large product now.
The core of the product is session recording, which I implemented using the rrweb library, which itself uses the Mutation Observer API. Everything else is pretty standard React/TypeScript/TanQuery for the front, Java/SpringBoot/PostgreSQL for the back.
No users, but working through the journey of trying to become a better copy writer / product manager / marketer! I've been getting some great advice on how to improve my landing page which has been exciting.
Looks great!
Have you checked Microsoft Clarifty - https://clarity.microsoft.com - it allows to generate heatmap and do session recording as well. And it is free. What are advantages of https://scryspell.com over it ?
I'm working on Habitat. It's a free and open source, self-hosted platform for communities to discover and discuss their local area. The plan is for it to be federated.
- The idea: https://carlnewton.github.io/posts/location-based-social-net...
- A build update and plan: https://carlnewton.github.io/posts/building-habitat/
- The repository: https://github.com/carlnewton/habitat
- The project board: https://github.com/users/carlnewton/projects/2
I'm building Concert Map [1], an interactive map of live music events all over the world.
I was frustrated with how difficult it was to find concerts and festivals when travel plans are flexible. Questions like:
- Is there any show I could catch on my next cross country road trip?
- I want an excuse to visit Seattle this summer, is there a good weekend I could pick where I could see live music there or along the way?
- I like to travel for one big music festival per year, is there a good match for me anywhere in the country, or abroad?
The idea behind concert map is to show all events on a map, and highlight the "most relevant" events for any area, so you can skim through the map and find events that are worth the travel. The "relevant" events depend on the view context. If you import your top artists, then the most relevant events are ones with artists that you listen to most. Otherwise, it's based on global artist popularity.
I'm currently working on integrating with more event providers, and working on loading performance. I've been using leaflet.js [2] which has worked great, but I'm experimenting with switching to supercluster [3].
[1] https://www.concert-map.com/
[2] https://leafletjs.com/
[3] https://github.com/mapbox/supercluster
Pretty cool! Looks like great project!
(https://casacompose.com) - Real Estate Listing Description Generator
I built it after a conversation with a friend thats a RE agent.
As an agent, he was literally spending 45+ minutes per listing just staring at a blank screen trying to make "3BR/2BA ranch" sound compelling.
The AI actually analyzes the photos to pull out key features he might miss and adds neighborhood context, like walkability score and gets professional descriptions in 30 seconds.
Just launched it as a standalone SaaS - would love feedback from anyone in the RE industry.
Looks like great idea!
Can be extended to other niches (e.g. anything that is sold on marketplaces, cars, boats, etc)
- Sharing my latest blog post on various socials to gather feedback - it's about the problems with pinning GitHub Actions by commit hashes: https://developerwithacat.com/blog/202508/github-actions-com...
- Casually investigating a construction tech startup idea with an architect
- thinking of what to do long term. Feeling pretty lost at the moment.
I contribute to the Happy Coder app https://github.com/slopus/happy (MIT License)
I've gotten a lot of use out of Claude Code in the past few months. The big problem for me is struggling to stay on top of multiple sessions in parallel. I know that all the cool kids on twitter have all these AI employees and agents farms... A lot of its probably bullshit, but directionally its clear that tooling will absorb some fraction of my workload.
I know I need to push myself to have 2 or 3 balls in the air. But without push notifications to know when Claude Code needs permissions or input, I end up "wasting" time a lot.
So having an app that sends discrete push notifications to my watch is pretty nice.
Oh, and before I forget, I'm pretty excited by the Voice Agent feature another contributor is working on https://x.com/giaccoangelo/status/1960550007272300964
You talk to a voice agent, then voice agent talks to Claude Code. The voice agent acts like a buffer for me to ramble and gather my thoughts. Then the voice agent writes the prompt for claude code when I'm ready. So that unlocked walk and code for me. Now as I walk to the train from the office, I can dump some of the brain ideas into Claude Code plan mode, and they'll be waiting for me in the morning.
Plus I've been using plan mode more and more.
I'm trying to change the nature of LLM compute, with an idea that's been scratching at the back of my mind for decades. If it works, someone (likely not me) will make Billions, or even Trillions of dollars. I'm very deep in eccentric old man/lone genius territory. Current odds favor the former.
BitGrid[1] - I'm trying to figure out how to do software for a BitGrid. Long, long ago I looked at computing, and decided that the best high performance strategy was something like an FPGA, but simplified, and optimized around overall performance, and not simply the lowest latency and most efficient use of silicon towards that goal. A vast 2d grid of 4x16 bit look up tables, with latches, seems to be optimal, to me. (It's a hunch, that I could very well be wrong about)
Getting an expression translated to a directed acyclic graph of binary logical operations seems to be something that's been achieved, with extensive help from ChatGPT5.[2] (Warning, it's a huge pile of stuff almost exclusively written by an LLM, not me, as I say at the start of the README.md)
The thing now is to figure out how to map that graph into a grid. The LLM has certain strong opinions about how things should go, using the phrase "physicalize" that have taken it off into a few rabbit holes I've gotten us back out of. I think I'll get us there, but not as soon as I had hoped.
The latest is that just getting bits to move across a grid took some backtracking, but forward progress is being made again. It seemed to think that you can just plop data into the middle of a grid (which required defeating the one advantage of a bitgrid... oops)
The thing I really want to know, is how big a BitGrid would it actually take to hold a frontier model LLM like ChatGPT5. Obviously it's not going to fit on an single chip, it's going to be distributed. I've actually managed to figure the distribution and communications part out. It's just the raw size and a good estimates for the current power requirements, and area of a cell implemented in 7nm or better silicon, and the math falls out the other side. You either save billions of dollars, and gigawatts, or you don't.
How? Imagine a grid of FPGAs big enough to hold ChatGPT5, with all the math programed directly into the logic, not separate ram/matmuls like a sane person would do. Add latching periodically through the flow so that it's effectively a 4,000,000 stage pipeline or so. If you then run it at 1 Ghz, you'd have a maximum throughput of 250 tokens/second. However, you could then run 3,999,999,999 other streams with the other slots through the pipeline. ;-)
I was hoping to get the above software figured out before the submission deadline for the TinyTapeout[3] Sky25a run in 18 days, to get an ASIC made. I think I might have to start working on that submission before I even know if it's worth it. My hunch, and every bit of math I've done before says it will be, but it would be nice to know for sure.
I hope to have some estimates figured out by next month, and will let you know if I made it into Sky25a or not.
[1] https://esolangs.org/wiki/Bitgrid
[2] https://github.com/mikewarot/Bitgrid_Python
[3] https://tinytapeout.com/
One of these again.. love to see all the projects, but man time flies when you're not doing much
Still slowly expanding https://wheretodrink.beer with new mini-features I want (like the obscurely hidden https://wheretodrink.beer/near-me - it probably need an interactive map), and adding on new venues. Struggling with getting any sort of organic traffic, but not sure that's even something I want..
Been toying with creating some other projects; one for helping with lost+found for SMBs in hospitality, and another with administrating "co-operatives" (collecting dues, coms, automating legal requirements).. So naturally you start building all the scaffolding that you could have bought, like gdpr-compliant privacy first multi tenant authentication.. That's fun, but I should know better.
https://codeinput.com - a tool to augment Git-related processes (merge conflicts, ownership, automation, etc..) I've been working on it for close to 10 months and it's still not to launch. On the other hand, I've been building it fully on Cloudflare Workers (WASM/Rust) and the whole back-end is a single wasm worker.