Iris - Self-Healing Observability Layer for Clinical AI | Google Rapid Agent Hackathon x Arize Track

Описание: IRIS is a self-healing observability layer for clinical AI. It evaluates every agent output in real time across 7 safety evaluators grounded in RxNorm and OpenFDA, traces everything to Arize Phoenix, and closes the loop autonomously:
an ADK agent reads IRIS's own observability data via the Phoenix MCP server
diagnoses the failure pattern, rewrites the system prompt
validates the fix against real captured failures
and deploys it - measurably improving the agent with no human intervention.

Built with Google ADK, Gemini 2.5 (Vertex AI), Arize Phoenix, and Cloud Run.

Live demo: https://iris-clinical-797220546460.us...
GitHub: https://github.com/adityashukla8/iris

Built by Aditya Shukla for the Google Cloud Rapid Agent Hackathon - Arize Phoenix Track.

#GoogleCloud #ArizePhoenix #GeminiAI #AIAgents #MCP