How databases work.

Without an index, a query reads every single row; with one, it jumps straight to the answer. See how a B-tree turns finding one row in a billion from a slow scan into a handful of hops — and why every index you add speeds reads but taxes writes.

A transfer debits one account and credits another — and a crash in between must never leave the money half-moved. Transactions give you all-or-nothing guarantees, and the write-ahead log is the trick that makes them survive a power cut.

Two people buy the last ticket at the same instant — who actually gets it? Isolation levels decide how concurrent transactions see each other's work, trading strictness for throughput to prevent lost updates, dirty reads and the double-sell.

Reads pile up until the database buckles under load. A cache keeps hot data in memory in front of it, absorbing most of the traffic so the database barely notices — as long as you handle invalidation and don't let the whole herd stampede the origin at once.

The finale jumps to the modern frontier: classic databases match exact values, but AI needs to search by meaning. Vector databases store embeddings and find nearest neighbors in high-dimensional space — the retrieval engine behind RAG and every "search that understands you" feature.