Starting a weekly blog

This is the first post, so it’s mostly a note to myself.

Most of what I pick up — from papers, blogs, source code — evaporates within a week. Writing it down helps it stick, and usually shows me the parts I didn’t really get. So I’m going to try posting once a week: not essays, not polished tutorials, just notes on whatever I was building or reading that week, written well enough that a version of me from a month ago could follow along.

what this will probably cover

• LLMs and NLP — the stuff I work on day to day.
• Retrieval — dense and sparse, rerankers, and the messy parts that never make it into the paper. I did my thesis on cross-lingual paragraph retrieval, so this is a soft spot.
• Agents and evaluation — I spend more time figuring out why an agent failed than celebrating when it works, and the failures are usually the more interesting story.
• Things I build for fun — like SmoLLM, the 109M-parameter model you can poke at on the projects page.

a few rules so this survives past week three

• Weekly, even when the post is short. Consistency over polish.
• Write about what I actually did, not what I think I should know.
• If I quote a number, it came from something I ran.

If any of it is useful to you, great. Mostly I’m writing it down so it stops evaporating.

See you next week.