Search used to mean matching keywords. If you searched “cheap laptop” and a page said “affordable notebook computer,” old search missed it. Embeddings are the technology that lets modern AI match on meaning instead of exact words.
The core idea
An embedding turns a piece of text (or an image, or audio) into a list of numbers — a “vector.” The trick is that this list captures meaning: texts that mean similar things get similar numbers, so they sit close together in a mathematical space. “Cheap laptop” and “affordable notebook computer” end up as near-neighbors, even with no shared words.
Why that is powerful
Once meaning is numbers, comparing meaning becomes measuring distance. This unlocks semantic search (find documents about a topic, not just documents containing a word), recommendations (find similar items), and retrieval for RAG systems (find the passages relevant to a question).
Where vector databases come in
Storing millions of embeddings and finding the closest matches in milliseconds is a specialized job. Vector databases — Pinecone, Weaviate, or the pgvector extension for PostgreSQL — do exactly that. When a RAG chatbot “looks up” relevant documents, it is doing a vector search under the hood.
Who should care
Anyone building search, recommendations, or AI assistants over their own content. Embeddings are the connective tissue of modern AI applications, and the concept is far simpler than the math suggests.