Vector Search

Vector Search converts all data (text, image, audio) into high-dimensional numerical vectors (embeddings) to power semantic and similarity-based retrieval.

Vector Search is the core engine for modern AI-driven retrieval: it transforms unstructured data (text, images) into high-dimensional vector embeddings, a numerical fingerprint of the content's meaning. Unlike traditional keyword search (lexical matching), this technique uses Approximate Nearest Neighbor (ANN) algorithms to calculate mathematical distance (e.g., cosine similarity) between a query vector and data vectors. Proximity in this vector space indicates semantic similarity, enabling systems to find what a user *means*, not just what they type. This capability is critical for applications like Retrieval Augmented Generation (RAG) in LLMs and high-accuracy recommendation systems (e.g., Spotify, Google Search); it demands specialized vector databases (Milvus, Pinecone) for billion-scale, low-latency performance.

https://www.ibm.com/topics/vector-search