Technology
Embeddings
Embeddings are numeric vectors (typically 768-1536 dimensions) that represent the semantic meaning of text. Two texts with similar meaning produce similar vectors. Used to power: similarity search, RAG, clustering, semantic classification. Generated by embedding models (OpenAI text-embedding-3, Voyage, Cohere).
More detail
If you embed the text 'How do I cancel my subscription?' and the text 'I want to stop paying', their vectors are close in space — semantic similarity, not lexical. This is the magic behind RAG: instead of keyword search, you find the most-similar chunks of your documentation to a user's natural-language question. Storage: vector databases (pgvector, Pinecone, Weaviate).