Hello, world, and why this site exists

This is the first post on the rebuilt version of my site. It's now a Next.js app, so I can write here without hand-editing HTML.

I mostly work on three things:

1. Deep learning from scratch: diffusion, JEPA world models, and LLMs reimplemented from the papers up in PyTorch.
2. Vision & perception systems: surveillance, captioning, and anomaly analysis over live video.
3. Full-stack & tooling: the apps and developer tools that put the models to work.

I'll use this blog to write up the things that don't fit in a README: the bugs that took a week, the ideas that didn't pan out, and the occasional one that did. Thanks for reading.