Engineering notes

I built a RAG endpoint for my portfolio and skipped the vector database

Retrieval is a fix for a corpus that doesn't fit. Mine is under three thousand tokens, so the retrieval step was pure cost — here's the reasoning, and what I built instead.