Embeddings

Embeddings are dense numeric representations (typically 384-3072 dimensions) where semantically similar items end up near each other in vector space. "King" and "queen" have similar embeddings; "king" and "banana" don't.

The operator-relevant use case is search. Embed your knowledge base. Embed the user's query. Find the closest matching chunks. That's the retrieval half of RAG, and it's how most modern AI search and recommendation systems actually work.

For most teams, the right move is to use an off-the-shelf embedding API (OpenAI, Voyage, Cohere) rather than training your own. The differentiation is in how you chunk documents, how you rerank results, and how you combine vector search with keyword search — not in the embedding model itself.