[2024] · Evals · Developer tools
[LLM eval harness]
[N]× faster prompt iteration
Problem
[Prompt changes shipped on vibes, and regressions reached users.]
What I built
[A golden set and graders wired into CI.]
Result
[Regressions caught, iteration speed.]
The AI decisions
[LLM-as-judge calibration, what you'd do differently.]