Glossary
Evals (Evaluation)
A repeatable test that measures how well an AI system performs on real examples, so you can improve it with evidence instead of vibes.
An eval is a test set for AI. You collect representative inputs with known good outcomes, then score the system against them, which turns 'it feels better' into a number you can track as you make changes.
Without evals, tuning an AI product is guesswork, and a change that helps one case can quietly break ten others. With them, you can compare models, prompts, and approaches and ship only what actually improves results.
How we use it
We build evaluations early and run them on every change, so you and we can see whether the system is getting better or just different.
Related terms
Get in touch
Want to put this into practice?
If this concept is relevant to something you're building, a short note is the fastest way to get practical help.