Embedding

An embedding is a fixed-length list of numbers that represents the meaning of a piece of content as a point in a vector space. A model maps text, an image, or audio into that space, and items with similar meaning land near each other, so similarity becomes a distance calculation rather than a keyword match. This is what lets a search for a laptop overheating fix return a document titled reducing thermal throttling in notebooks. Embeddings power semantic search, deduplication, clustering, recommendation, and classification, and they form the retrieval half of retrieval-augmented generation, where relevant passages are pulled in before the model answers. In practice a pipeline chunks documents, embeds each chunk with a model such as OpenAI's text-embedding family or an open Sentence-Transformers model, and stores the vectors in pgvector, Pinecone, or Qdrant behind an approximate nearest-neighbour index. Cosine similarity is the usual comparison. Two constraints matter. Vectors from different models are not comparable, so changing the embedding model means re-embedding the whole corpus. And semantic closeness is not correctness, since retrieval returns what is similar rather than what is true.