Embeddings
Dense numerical vector representations of text, images, or other data that capture semantic meaning in a form machines can compare.
What Is Embeddings?
Embeddings are fixed-length numerical vectors produced by an encoder model that represent the semantic meaning of input data. Two pieces of text that mean the same thing — "the cat sat on the mat" and "a feline rested on the rug" — will have embeddings that are close together in vector space, even though no words match. This mathematical representation of meaning is the foundation of modern semantic search, recommendation, and retrieval systems.
Text embedding models (OpenAI text-embedding-3, Cohere embed, BGE, Nomic) map text to vectors typically ranging from 256 to 3072 dimensions. Multimodal embedding models (CLIP, ImageBind) embed images and text into the same vector space, enabling cross-modal search.
Why It Matters
Embeddings unlock a class of AI capabilities impossible with keyword search: find conceptually related documents, detect duplicate content, cluster similar feedback, and retrieve context for RAG pipelines. They are the "universal language" that lets AI systems compare and organise heterogeneous data.
Real-World Examples
Semantic Search Engine
Embed all blog posts. A user searches for "reduce churn" and finds articles about customer retention — no keyword overlap required.
Duplicate Detection
Embed support tickets as vectors. Cluster nearby vectors to automatically group duplicate issues and route them to the same resolution.