Vector Database
Domain: LLM Concepts
🎯 Motivation
RAG needs to store and search millions of documents by meaning. Vector databases store text embeddings and find the most similar ones quickly.
📋 Concrete Example
Store 1 million document chunks as embeddings. When queried, compute the query embedding, find the 5 most similar document embeddings (cosine similarity), retrieve those chunks.
🔗 Analogy
A library catalogued by topic, not title - you search photosynthesis and it finds all books mentioning plant biology, even if the title doesn’t contain the word. Vector databases search by meaning.