evaluerBench
View on GitHub ↗The Need for Telemetry
When working with Local LLMs, understanding why an agent failed is just as important as knowing that it failed. I built evaluerBench to parse, visualize, and compare complex AI evaluation data in real time.
Architectural Choices
The dashboard focuses on high-density information display. By adopting an LCARS-inspired matrix grid and detailed side-by-side comparison views, I can quickly scan pass/fail rates, identify timeout regressions, and even launch fully interactive HTML/JS artifacts generated by the model directly within an iframe sandbox.
Tech Stack
- Framework: Astro
- Styling: Vanilla CSS (CSS Variables, Flexbox/Grid)
- Data: JSON serialized telemetry blobs