Practiced one out loud? Paste your answer into the free self-check and get a score in 30 seconds
What the interviewer is testing
Whether you know the blind spots of pure vector search, can explain how lexical and dense results are merged, and know when hybrid is worth the extra complexity.
Answer framework
- Define it: run lexical search (usually BM25) and dense vector search on the same query, then merge the results.
- Merge: reciprocal rank fusion (RRF) combines ranked lists without needing comparable scores.
- When to use it: queries with exact identifiers, product codes, names, error messages or rare domain terms.
- Default: it is a sensible default for enterprise and support search.
- Follow with a re-ranker, and measure recall against each method alone.
Sample answer
Hybrid search runs two retrievers over the same corpus. A lexical retriever such as BM25 scores documents by exact term matches, weighted by how rare the terms are. A dense retriever compares embeddings, so it matches meaning even with different wording. Each one fails where the other is strong: embeddings tend to blur an error code like E1047 or a part number into similar-looking strings, while BM25 misses a paraphrase that shares no words with the query.
To combine them I usually use reciprocal rank fusion: each document gets the sum of 1 / (k + rank) across the two lists, with k often around 60. It works on ranks, so I do not have to calibrate BM25 scores against cosine similarities.
I use hybrid search whenever users type identifiers, names, version numbers or jargon, which covers most enterprise, support and developer search. Then I put a re-ranker on the merged candidates and check on an eval set that hybrid actually beats either retriever alone.
Likely follow-ups
Common mistakes
- Assuming vector search alone handles exact identifiers well.
- Adding raw BM25 and cosine scores together without normalization.
- Skipping evaluation of each retriever on its own.
- Forgetting that BM25 needs the same text preprocessing at index and query time.
The 0 to 1 AI Engineer Interview Playbook covers the answer frameworks, system-design walkthroughs and an 8-week plan.
Get the playbook 40% off with code EARLYBF until Nov 3, 11:59 PM PTEARLYBF: 40% off until Nov 3Smaller start: Score your answers against 50 rubrics, $9. 50 AI engineer questions, one rubric each, PDF on Gumroad.
Interviewing for PM or TPM? See the full PM package, $59 on Gumroad: a resume system, a PM, TPM and PGM interview handbook, a 1:1 cheat sheet and a PM playbook. PM and TPM content, not AI engineer prep.