Vinícius Raposo

Applied AI Reliability Engineer

AI systems you can leave on without babysitting.

Profile pin · Python

evalforge

Regression gate for RAG and agents that catches answers which look right but cite the wrong evidence, with pass / fail / indeterminate release exits.

358 Python + 36 frontend tests. The bundled corpora are synthetic and agent-authored, with no human validation, so the demo shows the gate working, not a model-quality result.

Language: Python Updated: 2026-08-13 Visibility: public
View on GitHub Back to catalog

Without regression evals on retrieval and conversation quality, silent drift ships to production—small model or prompt changes break answers and nobody notices until users complain.

Quality drift is caught before it ships.

Offline demo:

cd frontend && npm run demo:ui

Walkthrough

Failed retrieval case with judge scores before ship.
Failed retrieval case with judge scores before ship. Sample data.
Regression suite flags retrieval, citation, and refusal drift together.
Regression suite flags retrieval, citation, and refusal drift together. Sample data.
Current run compared against baseline to catch regressions.
Current run compared against baseline to catch regressions. Sample data.

Why it matters

Regression gate for RAG and agents that catches answers which look right but cite the wrong evidence, with pass / fail / indeterminate release exits.

Repo facts

Default branch: main. Stars: 0. Forks: 0. Created: 2026-05-31.