Folio V · Notes
Marginalia from real builds.
Short, technical and measured. What worked, what didn't, and the numbers.
- No. 02What a real citation is in a RAG system (and why most chatbots fake it)A citation isn't a link at the bottom of an answer. It's a verifiable chain from each claim back to a specific passage. Here's how to build citations your users can actually trust.
- No. 01How I cut a RAG system's response time from 4.2s to 1.2sThe bottleneck was retrieval, not the LLM. Three changes (512-token chunks, hybrid search and an embedding cache) took a FastAPI RAG system from 4.2s to 1.2s and context accuracy from 71% to 89%.