Vector Databases
The Library Sorted by Vibes
7 min read
The Library Sorted by Vibes
A normal database finds exact matches. A vector database finds things that feel similar — even with completely different words.
Search "affordable cars in Mumbai" in a normal database and it finds those exact words. Search in a vector database and it finds results about "budget vehicles in Maharashtra", "low-cost hatchbacks near Pune", and "economic transport options in metro cities" — because they're semantically close. Vector databases store embeddings — meaning as numbers — and find neighbours by similarity, not keywords.
In Plain English
A vector database stores embeddings (meaning as numbers) and retrieves the most semantically similar items to a query — not just exact keyword matches. It's the storage layer that makes RAG and semantic search possible.
The Technical Picture
Vector databases (Pinecone, Weaviate, Qdrant, Supabase pgvector) store high-dimensional embedding vectors and support approximate nearest-neighbour (ANN) search using algorithms like HNSW or IVF. They enable sub-millisecond semantic retrieval across millions of vectors at scale.
Real-World Examples
- Pinecone powering the retrieval layer of an enterprise RAG chatbot
- Supabase pgvector enabling semantic search in a Next.js application
- Notion AI using vector search to find relevant notes across your entire workspace
Vector databases find meaning-neighbours, not keyword-matches — that's what makes semantic search feel magical.