What is hybrid search, and why combine keyword and vector search?
Hybrid search runs keyword search such as BM25 and vector search together and merges the rankings. Vector search catches meaning and paraphrases, while keywords catch exact terms like product codes, names and rare jargon that embeddings can miss.
Explain it simply
Searching a company's help centre shows why. A user types "ERR_4012". Embeddings may not treat that exact code as special, but a keyword match finds it instantly. Another user types "can't log in after changing phone", and a keyword search misses the article titled "Device change and account access", while embeddings find it.
Hybrid search uses both. The system retrieves candidates from each method, then fuses the rankings, for example with reciprocal rank fusion, which scores items by their position in each list, or with a weighted score.
It is particularly useful in domains full of identifiers, abbreviations, names and legal or medical terms. The cost is running two indexes and tuning how to combine them. Measure on your own queries before assuming it helps.
Example
Query: "ERR_4012 on UPI payment"
Keyword list: [doc7, doc2, doc9]
Vector list: [doc2, doc5, doc7]
Fused ranking puts doc7 and doc2 at the top.What to say in the interview
"Keyword search is strong on exact tokens, vector search on semantics. Hybrid retrieval merges both, commonly via rank fusion, which improves recall for queries with identifiers or jargon. I would validate the gain on a labelled query set."