← Back to Projects

open-evals

Provider-independent evals and prompt self-improvement platform

open-evals - Image 1
open-evals - Image 2
open-evals - Image 3

Context

Built open-evals as an MIT-licensed platform for eval rows, failure attribution, calibration, prompt optimization, and release decisions. It imports traces as evidence for evals without turning into a generic observability product.

SCOPE

  • Deterministic evals that work without AI credentials
  • Trace import adapters for common eval and observability formats
  • Prompt release, optimization, A/B/n assignment, and privacy-aware artifact handling

MY ROLE

  • Product Builder
  • AI Engineer

TOOLS

  • Evals
  • Prompt Optimization
  • TypeScript
  • Python
  • PostgreSQL

TIMELINE

Jun 2026

LINKS