Working notes on AI engineering — retrieval, embeddings, and the parts of applied machine learning that only make sense once you’ve shipped them.
Notes here are written for my future self: short, opinionated, and biased toward the detail that actually mattered in practice. They get revised as my understanding does.
Start here
- Late chunking — pool after attention, not before
How this is organised
Notes carry a track (APPLIED for things I’ve built with, THEORY for things I’m
still reading into), a topic, and prereqs where the note assumes something earlier.
Use the Explorer on the left to browse, Search to jump straight to a term, or
the Graph on the right to follow how notes connect.
Some notes are private and ask for a password.