What is hybrid search?
Combining multiple retrieval signals, commonly dense semantic search and sparse keyword search.
Interview Questions: Hybrid Search and Query Routing
Use these to explain hybrid retrieval as a controlled architecture decision.
When is dense retrieval useful?
Paraphrase, concept search, vague user language, and semantic similarity.
When is sparse retrieval useful?
Exact IDs, error codes, legal clauses, SKUs, names, and rare tokens.
What is reciprocal rank fusion?
A rank-based method for merging results from different retrievers without relying on incompatible raw scores.
What is query routing?
Choosing a retrieval path based on intent, exact-term need, domain, risk, latency, and cost.
Why not use hybrid for everything?
It can add latency, cost, complexity, duplicate handling, and fusion risk without improving quality for simple queries.
What should a route trace include?
Route, reason, retrievers used, filters, top-k, fusion method, selected chunks, and fallback behavior.
How do you evaluate hybrid retrieval?
Measure route-specific Recall@k, exact-term recall, MRR, citation support, duplicate rate, latency, and failure labels.
What is a common hybrid failure?
Combining raw dense and sparse scores without normalization, causing one retriever to dominate incorrectly.
How should high-risk queries be routed?
Use stricter filters, exact-term protection, higher recall paths, reranking, citation checks, and trace review.