Database systems specialised for storing high-dimensional vector embeddings and performing similarity search over them, typically using approximate nearest neighbour indexing algorithms such as HNSW and IVF.
Semantic Classification
Content
- Vector databases store embeddings, which are dense numeric representations of text, images or other data produced by machine learning models. They index these vectors so that, given a query vector, the system can efficiently return the most similar stored items using distance or similarity metrics.
- To scale to large collections they rely on approximate nearest neighbour algorithms that trade a small loss of exactness for substantial speed gains. Vector databases are a core component of semantic search and retrieval-augmented generation pipelines.