How it works
An embedding model takes a piece of content and outputs a fixed-size vector (typically 768-3072 dimensions). Similar content produces similar vectors. Cosine similarity or Euclidean distance measures how close two embeddings are in meaning.
Example
The phrases "How do I cancel my subscription" and "I need to end my account" produce similar embeddings even though they share few words, allowing a RAG system to retrieve the same help article for both.