What is Vector Indexing

Definition

Vector indexing is the technique of organising embedding vectors into specialised index structures that enable fast approximate nearest-neighbour search, so similarity queries over millions of vectors return relevant results in milliseconds rather than scanning every vector.

« Back to Glossary Index

Enables fast similarity search over millions of embeddings
Uses approximate nearest-neighbour structures to avoid full scans
Trades a little accuracy for large gains in query speed
Underpins scalable semantic search and retrieval

Real World Example

A search service builds a vector index over its document embeddings so a query is matched against millions of vectors in milliseconds, returning semantically relevant results without comparing the query to every vector individually.

FAQs

Why is vector indexing needed?

Comparing a query against every vector is too slow at scale, so an index enables fast approximate nearest-neighbour search.

What index structures are used?

Approaches like HNSW graphs and IVF partitioning organise vectors for efficient similarity search.

What is the accuracy trade-off?

Approximate indexes return slightly less exact results in exchange for dramatically faster queries, tunable by parameters.