Long-form & notes
Writing
Notes on machine learning systems, evaluation, and operating models in production.
-
Build notes #1: My AI agent beta-tests its own memory
Building Muninn, a local LLM librarian for Markdown vaults, and having Claude Code beta-test its memory and log bugs into the vault itself.
-
Six Years of Consulting and One Bad Tkinter App
Six years of consulting on forecasting and machine learning, alongside engineering jobs, team leadership, and an MSc thesis still in progress.
-
How Linux Changed the Way I Think
Years of breaking and fixing Linux left me with an instinct: most things on a computer are not fixed objects, but systems you can question and bend.
-
Small Example of Real Agentic Benchmarks
Test AI on rare, undocumented tools (like the Soupault static site generator) rather than common ones like React. This forces the AI to infer logic from provided examples rather than relying on its training data.