Hello world: why this site exists
I’ve always learned best by rebuilding things until they click. This site is where I do that in the open.
The idea is simple: pick something I want to understand, build a small version of it from first principles, and write up what actually happened, including the parts that broke. Working in public keeps me honest, and hopefully some of it helps someone walking the same path.
What I’ll share here
- Machine learning, mostly from first principles: training models, breaking them, and figuring out why they behave the way they do.
- Frontier LLM research: reading the papers, reproducing what I can at small scale, and probing how large language models actually reason and where they fail.
- Reinforcement learning: agents that learn from feedback, from the classic toy problems up to the messier modern setups.
- Interactive experiments: small things you can play with right in the browser, like the ones under Experiments. If I can make an idea tangible, I will.
Expect short, honest write-ups over polished essays. Build a thing, learn something, write it down, move onto the next.
If you want to follow along, everything lands in the RSS feed.