Self-hosted LLM evaluation framework

Shipped a Phoenix-based evaluation framework in three days, giving engineers self-hosted inspection of model output, reasoning steps and tool-call traces — no external cloud hosting, so internal iteration stayed fast.

AI EvalsDeveloper ToolingPythonLLM Integration

Self-hosted LLM evaluation framework

I shipped this Phoenix-based evaluation framework in three days. It gave engineers a self-hosted way to inspect model output, reasoning steps, and tool-call traces.

Delivery constraints

The framework did not use external cloud hosting. Keeping it self-hosted supported fast internal iteration.