Vector Embedding

Vector Embedding transforms unstructured data (text, images, audio) into high-dimensional numerical arrays (vectors), enabling machine learning models to process semantic meaning.

Vector Embedding is the core process: it maps complex, unstructured data—like a document or an image—to a dense, continuous vector space. This numerical array, often 1,536 or 3,072 dimensions in models like OpenAI’s `text-embedding-3-small`, mathematically encodes the data’s semantic meaning. The key is proximity: vectors for similar concepts cluster tightly, allowing algorithms to quickly calculate relatedness using metrics like cosine similarity. This foundational capability powers modern semantic search, Retrieval-Augmented Generation (RAG) applications, and recommendation systems, moving beyond simple keyword matching to true contextual understanding.

https://openai.com/docs/guides/embeddings/