Also known as: vector embedding
In plain English
An embedding turns words into numbers in a way that captures meaning. "Dog" and "puppy" get similar numbers, while "dog" and "invoice" don't. This lets computers search by meaning, not just exact words.
In practice
Embeddings power semantic search, recommendations, duplicate detection and RAG. If a user searches "laptop won't turn on", embedding search can find an article titled "Device fails to power up".
Under the hood
An embedding model maps input to a dense vector, often hundreds to a few thousand dimensions, trained so that semantic similarity corresponds to vector similarity, usually measured with cosine similarity. Embeddings from different models are not interchangeable.
Example
"We created embeddings for 10,000 knowledge articles to improve search."