Good catch—it's actually a hybrid approach:
1. Anomaly Detection & Visualisation: This runs 100% locally via WASM using the linear-time algorithms I shared here previously.
2. Forecasting: This hits our DriftMind API [1][2].
We designed the API to be transient-only. The data is processed in-memory for the forecast and immediately discarded; absolutely nothing is persisted to disk or used for training. We wanted to keep the browser lightweight while still offering the heavy-duty forecasting from our core engine
I remember your post from a few weeks ago about the linear-time dimensionality reduction. Is this tool using that same algorithm under the hood?
I like the focus on privacy, but I'm curious about the 'Forecasting' feature. Is that also running entirely in WASM? I'd imagine running a decent forecasting model in the browser might be heavy on the client side.
Good catch—it's actually a hybrid approach: 1. Anomaly Detection & Visualisation: This runs 100% locally via WASM using the linear-time algorithms I shared here previously. 2. Forecasting: This hits our DriftMind API [1][2]. We designed the API to be transient-only. The data is processed in-memory for the forecast and immediately discarded; absolutely nothing is persisted to disk or used for training. We wanted to keep the browser lightweight while still offering the heavy-duty forecasting from our core engine
[1] https://thingbook.io/doc/devGuide.html [2] https://medium.com/towards-artificial-intelligence/reflexive...
I previously shared my linear-time dimensionality reduction algorithm https://news.ycombinator.com/item?id=46285535 I've now wrapped that + my one-pass forecasting engine into a free tool for non-coders. Tool explained here: https://medium.com/@roman.f/why-were-liberating-anomaly-dete...
I remember your post from a few weeks ago about the linear-time dimensionality reduction. Is this tool using that same algorithm under the hood? I like the focus on privacy, but I'm curious about the 'Forecasting' feature. Is that also running entirely in WASM? I'd imagine running a decent forecasting model in the browser might be heavy on the client side.