Welcome to Vector Databases, the sixth module of the Generative AI & LLM Engineering course. In Module 5 you built a semantic search engine by embedding a handful of documents and comparing every one of them to each query. That works beautifully for fifteen FAQs — but what happens at fifteen thousand, or fifteen million? Re-embedding and re-scanning everything on every query is too slow and wastes memory. A vector database solves this: it stores your embeddings, indexes them for fast nearest-neighbor search, and lets you filter by metadata — all while persisting to disk so you embed your data once and query it forever.

You’ll start with why a dedicated store is needed and what an index actually buys you. Then you’ll get started with Chroma, an open-source vector database that runs locally and for free — creating a collection, adding documents, and querying by meaning in just a few lines. You’ll add metadata and filtering so you can search within a category, and make the database persistent so it survives restarts. The capstone wires it all together into a searchable knowledge base you can query like a real product.

Every example runs for real on your machine — no API key and no cost, because Chroma embeds and searches locally. By the end you’ll have the storage and retrieval layer that powers retrieval-augmented generation, the very next module, where these searches feed real context to Claude.

Start with Lesson 1, where you’ll see exactly why brute-force search breaks down and what a vector database does about it.