Retrieval & RAGintermediate11 min · 9 steps
Hybrid Search
Run keyword and vector search side by side and fuse their rankings, so exact identifiers and paraphrases both make the top-k.
requesterrordata
ProblemStep 1 of 9
Dense search can miss the one word that matters
A user asks: “Configure NVIDIA_VISIBLE_DEVICES”. The doc that answers it contains that exact string — but vector search ranks it poorly.
The identifier tokenizes into odd fragments, the embedding captures meaning rather than exact strings, and the term may be rare in the embedder's training data. On technical docs, version numbers and function names can carry most of the information.
Watch outCodes, acronyms, product names and error numbers are exactly the queries users can't paraphrase — and exactly where dense-only retrieval is weakest.
Lesson 5 of 37 in Retrieval & RAG