Vector Database
A database designed to store and search high-dimensional numerical representations (vectors) of text, documents, or other data. Vector databases power semantic search and retrieval-augmented generation (RAG) by finding items that are conceptually similar — not just textually identical.
What is a Vector Database?
When an AI model processes text, it converts words and sentences into vectors — lists of numbers that encode meaning. Two sentences that say the same thing in different words will produce similar vectors. A vector database stores these numerical representations and enables fast similarity search: given a query, find the stored items whose vectors are closest.
Traditional databases answer questions like "find all records where supplier_name = 'Acme'". A vector database answers questions like "find all documents that are semantically similar to this query about late delivery penalties" — even if those documents never use the words "late," "delivery," or "penalty" explicitly.
How Vector Databases Work
The process has three steps. First, documents are converted to vectors using an embedding model and stored in the vector database. Second, when a query arrives, it is converted to a vector using the same model. Third, the database runs an approximate nearest-neighbor search to return the most similar stored items — typically in milliseconds, even across millions of documents.
Common vector databases: Pinecone, Weaviate, Qdrant, pgvector (PostgreSQL extension)
Key metric: Cosine similarity or dot product — measures how close two vectors are in high-dimensional space
Used with RAG: Retrieve relevant document chunks, pass them to an LLM as context for answering questions accurately
Vector Databases in Operations
For operations teams, vector databases enable AI systems to search institutional knowledge that was previously un-searchable. Imagine a procurement team that needs to check whether a new supplier contract clause conflicts with company policy. With a vector database indexing all past contracts and policy documents, the AI can retrieve the three most relevant precedents in under a second — regardless of how the clause is phrased. At Lleverage, vector search is used to match incoming documents against known templates, find historical exceptions similar to a current anomaly, and surface relevant policy text during automated decision-making.