Listen to this lesson
Unlock audio and more
Audio streaming, downloadable PDFs and certificates come with Plus and Pro.
What it means
Keyword search matches characters. Semantic search matches meaning. Each fails precisely where the other succeeds: keyword search misses a document about "terminating your subscription" when the user typed "how do I cancel", and semantic search returns something plausibly related when the user typed an exact order number.
Hybrid search runs both and combines the rankings, so an exact identifier is found by the keyword half while a differently-worded question is found by the semantic half. The merging step is usually a rank-fusion method that blends two ordered lists without needing the two systems' scores to be comparable — which they are not.
Most managed search and vector platforms now offer this as a built-in mode rather than something you assemble.
Why it matters
This is the single most common fix for a retrieval system that demos beautifully and disappoints in production. Demo queries are conversational, which flatters semantic search; real queries contain part numbers, error codes, customer names and ticket references, which it handles badly.
In practice
Default to hybrid rather than pure vector search. If you have already shipped semantic-only and users complain that search "cannot find things it obviously has", check whether the failing queries contain identifiers before touching the embedding model.
Where this shows up
Tools and models in our catalog.
PineconeThe leading managed vector database for AI applications. Serverless pricing, 99.99% SLA, and billions of vectors at millisecond query speeds. Widely used in production RAG systems.
Elastic AI SearchESRE (Elasticsearch Relevance Engine) for retrieval-augmented generation. Hybrid vector + keyword search combining traditional full-text with semantic vector search in a single platform.
Supabase VectorPostgreSQL-based vector storage using the pgvector extension. Seamlessly combines traditional relational data with vector search in a single database.