An embedding turns meaning into numbers so a computer can measure similarity.
DefinitionWhat it means
An embedding is a dense numerical vector, typically hundreds or thousands of dimensions, that represents the semantic meaning of a piece of text, image, or other data, produced by a trained embedding model. Items with similar meaning end up with vectors that are close together in that high-dimensional space, which lets a computer measure similarity mathematically instead of matching exact keywords.
Why it mattersWhy you should care
Embeddings are the foundation of semantic search, recommendation systems, and retrieval-augmented generation, since they let a system find conceptually relevant results even when the wording differs completely from the query. Choosing an embedding model, and keeping it consistent between indexing and querying, is one of the most common and costly mistakes teams make when building a retrieval system.
At a glanceSee it
Similarity is not magic — it is the cosine of the angle between two vectors, so meanings that point the same way score as alike.
A retrieval pipeline embeds document chunks into an index and embeds the question the same way, then returns the nearest chunks — the dashed note shows how bad chunk sizing quietly breaks it.
Where you see itIn the wild
- Semantic search features returning conceptually related results
- RAG pipelines embedding documents before storage
- Recommendation systems matching users to similar items