Glossary
Embeddings
Also: Vector search
Number sequences that represent the meaning of a text. Similar content sits close together in vector space, even without shared words.
This allows searching by meaning instead of by string: a query about notice periods also finds a paragraph that talks about end of contract.
The vectors are stored in a database with a suitable index. Whether that needs a dedicated product or an extension of your existing database depends on volume, not on principle.
How you notice it
- Users search by meaning, not by exact words.
- Full text search only finds the right thing with the right phrasing.
- There are many documents but no useful ranking.
Frequently asked
Which embedding model should I pick?
One that knows your language and your terminology, and then the same one permanently. Switching means recomputing the entire corpus, because old and new vectors are not comparable. That is why the model name belongs in the store, not just in the code.