Hybrid search (RAG)
Combining vector search (semantic similarity) with classic keyword search (BM25/lexical) in a RAG pipeline, and merging the results. It captures both "means the same thing" and "contains this exact word" — the best of both worlds.
Problem: pure vector search is good at meaning but can miss exact terms (product names, codes, rare words, acronyms); pure keyword search catches exact matches but misses synonyms and rephrasings. Hybrid runs both in parallel and combines the rankings, often with Reciprocal Rank Fusion (RRF), which merges two hit lists without having to calibrate their scores against each other. Win: noticeably better accuracy, especially for domains with technical terms, IDs, and code. Often followed by a reranker that fine-tunes the final order. One of the most reliable improvements over a basic RAG. Related to reranker and RAG.