hybrid retrieval

Why pick just one? Hybrid retrieval runs a dense embedding search and a sparse keyword search side by side and then merges their results, so you get semantic recall and exact-match precision in a single pass. The paraphrased passage that dense search loves and the exact product code that keyword search guarantees both make it into the candidate set.

Mechanically, each retriever returns its own ranked list, and the lists are fused into one — often with reciprocal rank fusion, which rewards passages ranked highly by either method, or with a weighted blend of the two scores. A reranker frequently follows to clean up the merged order. In practice hybrid retrieval lifts recall noticeably over either method alone, which is why it has become a common default for production RAG.