Vector Search

Vector search, also known as similarity search or nearest-neighbour search, is a method of finding items that are semantically similar to a query by comparing their vector representations. Items (documents, images, products, or any data) are converted into numerical vectors using embedding models, and search is performed by finding the vectors closest to the query vector. The closeness between vectors is measured using distance metrics such as cosine similarity, Euclidean distance, or dot product. Vectors that point in similar directions (high cosine similarity) represent semantically similar content. This allows vector search to find relevant results even when the exact words differ from the query. For practical applications with millions or billions of vectors, exact nearest-neighbour search is too slow. Approximate nearest-neighbour (ANN) algorithms — such as HNSW, IVF, and ScaNN — trade a small amount of accuracy for dramatically faster search times, making vector search feasible at scale.

Vector search enables applications that understand meaning, not just keywords. A customer searching for 'comfortable running shoes for beginners' can find relevant products even if none of those exact words appear in product descriptions. A support agent searching for 'customer cannot access their account' can find relevant knowledge base articles about login issues, password resets, and account lockouts. This semantic understanding dramatically improves search quality across enterprise applications. Internal knowledge search, product discovery, customer support matching, content recommendations, and duplicate detection all benefit from vector search's ability to match by meaning. Vector search is the core retrieval mechanism in RAG systems, making it foundational to most enterprise AI applications. Investing in high-quality vector search infrastructure — including good embedding models, efficient indexing, and appropriate distance metrics — directly impacts the quality of AI-powered applications.