Recall which retrieval tactic best addresses exact terms, semantic mismatch, fusion, final ordering, and global synthesis.
Exact vs semantic BM25 or dense retrieval? Most enterprise RAG stacks need both. Fusion Can we just average BM25 and vector scores? Use when combining heterogeneous retrievers. Your line Use rank fusion first; the raw scores do not share a calibrated scale. Score averaging creates false precision and lets one scale dominate. RRF combines positions without pretending scores mean the same thing. Reranking When do you add a reranker? After first-stage retrieval has broad enough recall and candidate count is small enough to spend extra compute. Rerankers are final-ordering tools, not substitutes for missing candidates. Global context Hierarchy or graph?…
Sign up free — one personalized lesson every day, matched to your role and goals.
Already have an account? Sign in