Retrieval Tuning Battlecards
Recall practical RAG checks during real work.
“Search is bad; increase top-k for everything.” Your line First decide whether this is a recall miss or a precision problem; top-k helps one and can hurt the other. More chunks can bury the governing source in noise. It uses precision and recall diagnosis before tuning. Top-k increase vs metadata filter Use filters for country, role, status, product, and access before asking the model to sort it out. Symptom: exact source exists but ranks below broad overviews. Try reranking, query expansion with domain terms, or authority weighting. The retrieval set has the answer; ranking is the problem. Symptom: answer cites…
Sign up free — one personalized lesson every day, matched to your role and goals.
Already have an account? Sign in