What an eval is, and how to build one
How do you know a prompt or model change made your AI feature better?
An eval is a test for an AI feature: real inputs, a right answer for each, and a check that marks pass or fail. Build it from your own failures and run it on every change. Then you'll see whether a change helped.
