What is Embeddings?

An embedding is a list of numbers (a vector) that represents the meaning of a piece of text in a high-dimensional mathematical space. Text with similar meaning produces vectors that are close together — 'buy running shoes' and 'purchase athletic footwear' will have nearly identical embedding vectors, even though they share no words.

Embeddings are generated by encoding models (OpenAI's text-embedding-3, Cohere's Embed, or open-source alternatives). Once generated, they're stored in a vector database where similarity searches can find the closest matches to any new query vector in milliseconds.

In production AI applications, embeddings power three key capabilities: semantic search (find meaning, not keywords), RAG (retrieve relevant documents for an LLM to reason over), and recommendation systems (find items similar to what a user liked). Almost every serious AI application involves embeddings somewhere.