Why it matters
Hybrid search combines keyword precision with semantic recall. Error codes, SKUs, and legal defined terms often need exact matches semantic search alone misses.
How it works
Run BM25 or equivalent plus vector search, fuse scores with weighted or reciprocal rank fusion, then apply governance filters.
Example
Nintendo hybrid search finds AUTH-4412 in API docs via keyword match while also retrieving conceptual authentication articles via embeddings for how do I fix login errors.
Common mistakes
- 1Semantic-only on technical docs with critical codes
- 2Arbitrary fusion weights never tuned on eval sets
- 3Keyword leg retrieving internal codenames customers never type
Definitions are useful. Governed knowledge is better.
Wiki helps you turn these concepts into a real system for the knowledge behind your AI.
See Wiki in action