About

I work on AI agent memory research and full-stack engineering. Books, podcasts, and ideas show up here too. This site is where I write about whatever earns the post.

Research: agent memory

A tiered memory system for AI agents. Per-session SQLite working memory on the device side, a Postgres-backed shared corpus on the team side, HDBSCAN clustering and a cross-encoder reranker as sidecars. Five-tier RRF recall, workspace scoping, and a dormant cognitive-primitives extension (ACT-R, Hebbian, Bayesian) that the next eval cycle will resurrect.

Currently sitting at R@5 = 97.4% on LongMemEval-S, ahead of every published reference. The interesting work is what comes after that ceiling: measuring memory the way human expertise actually behaves (decay, co-activation, surprise weighting) rather than as flat similarity retrieval.

Day-job: infrastructure and full-stack

The rest of the stack, written about with the same honesty. Topics:

• Cluster ops and GitOps. What broke this week, what survived me.
• Observability and instrumentation. The alerts that mean something, and the ones I muted three months ago.
• Backend and frontend engineering. Small interfaces, errors as values, opinions I'll soften when something burns.
• Databases at scale. The query that works in dev and tanks in prod.
• Full-stack work. Boring on the resume, load-bearing in the codebase.
• AI agent tooling in daily use. What saves time vs what feels productive but isn't.

Elsewhere

Posts about anything I want to write about.