AgentEval

Co-built an evaluation platform measuring A2A agent accuracy and consistency across repeated calls in network-restricted Daytona sandboxes, with DeepEval scoring and semantic response clustering.

What I built

1. 01

   A Next.js and FastAPI platform that measures A2A agent accuracy and consistency across repeated calls.

2. 02

   Each evaluation question runs in its own network-restricted Daytona sandbox.

3. 03

   DeepEval scoring combined with semantic response clustering to detect consistency drift.

4. 04

   Per-run scores and explanations in an interactive dashboard.

Stack

• Next.js
• FastAPI
• DeepEval
• A2A
• Daytona