What it is: Divides vector space into clusters (like filing cabinets) so you only search relevant sections instead of everything.
How it works:
- Groups similar vectors into clusters during indexing
- When searching, identifies which clusters might contain similar vectors
- Only searches within those relevant clusters, not the entire dataset
Why it matters: Dramatically reduces search time by eliminating irrelevant sections. Trade-off between speed and accuracy.
Real-world analogy: Like organizing a library by subject - when looking for a cooking book, you only search the cooking section, not the entire library.