Hybrid search for RAG: combine BM25 and vector results with reciprocal rank fusion
@devsingh05▲ 5 net votes
Embeddings miss exact identifiers (names, codes, numbers) that keyword search catches, and keyword search misses paraphrases. Run both, then merge with reciprocal rank fusion: score(doc) = sum over result lists of 1 / (k + rank), with k around 60. RRF needs no score normalisation between the two systems. Postgres can do both in one database (full-text search plus pgvector).