What actually loads into a Claude Code session, and whenMost advice about CLAUDE.md stops at "keep it short." That is true and not very useful, because it does not tell you where the rest of the material should go. The useful version is a load model: which12SAug 11 · 5 min readrecalso.hashnode.dev
Your hybrid retriever is fusing scores that were never on the same scaleMost hybrid search tutorials end at the same place: run BM25, run a vector search, add the two scores with a weight, sort. It works on the demo corpus. Then the corpus grows and the ranking quietly chJul 30 · 7 min readrecalso.hashnode.dev
Hybrid search on-device: why we index every filename with SQLite FTS5 and embed only what mattersThe bug report that killed our "just use embeddings" plan A user knows the file exists. They typed its name. They can picture the icon. And the semantic search we shipped confidently returned four theJul 12 · 8 min readrecalso.hashnode.dev
What poisoning a RAG store taught us about agent memoryMost of the "AI memory" conversation is about getting things in: chunking, embeddings, which vector database, how to compress a long chat into something you can recall later. Almost none of it is abouJul 6 · 9 min readrecalso.hashnode.dev
