ADVANCED PROJECT
Regression eval platform for prompts
Quick answer
Client brief: A team changes prompts weekly and has no way to see what broke.
Deliverables
- 1. Versioned prompts and datasets
- 2. A CI check that blocks regressions
- 3. A results dashboard
Rubric: it passes when
- ☐ A bad prompt change fails CI
- ☐ Results are reproducible
- ☐ Datasets are versioned