Vector databases and indexes

Exact nearest-neighbour search does not scale, so vector indexes trade a little recall for a lot of speed. Knowing the trade-off keeps you out of trouble.

What this lesson covers

• Understand approximate nearest neighbour search
• Trade recall against latency deliberately
• Store metadata alongside vectors for filtering

This lesson sits in Embeddings and Vector Search, part of Retrieval-Augmented Generation from Scratch. It assumes what came before it and leads directly into the next lesson in the module.