Retrieval & Agents
Hybrid Search
Combining keyword search with meaning-based vector search to get the strengths of both.
Hybrid search blends classic keyword matching, which nails exact terms, names, and codes, with vector-based semantic search, which captures meaning even when the words differ. Merging their results covers cases where either alone would miss, which is why it is a common default in production retrieval. The two score sets are combined with a weighting or a fusion method.
Related terms
Semantic Search
Search that matches on meaning rather than exact keywords, so related ideas surface even with different wording.
Vector Database
A database built to store embeddings and quickly find the items most similar in meaning to a query.
Retrieval-Augmented Generation
A technique that looks up relevant documents and feeds them to a model so its answers are grounded in real sources.
Reranking
A second pass that reorders retrieved results by relevance, surfacing the best few before they reach the model.