Agent-powered helpdesk ticket routing
“Evaluation shows high routing accuracy (e.g., 4/5 correct); generates dashboard-ready results.json with full reasoning logs.”
solo-developer
Why they built it
Customer support teams receive large volumes of tickets and manually reviewing and routing them delays resolution and increases operational costs.
What worked
Multi-agent specialization, flexible model swapping via OpenRouter, evaluation framework scoring routing accuracy, interactive Streamlit dashboard.
What broke or was painful
OpenAI API deprecations fixed by OpenRouter switch; Pydantic integration issues via custom mock; noisy LLM outputs parsed for clean JSON; API rate limits managed with multiple keys.
The result
Evaluation shows high routing accuracy (e.g., 4/5 correct); generates dashboard-ready results.json with full reasoning logs.