Guardian AI

Inspiration

Every year, disasters claim thousands of lives while emergency response teams struggle with coordination delays and slow decision-making. We witnessed precious minutes lost between detection and evacuation during recent emergencies. What if AI agents could coordinate emergency response faster than any human team? This inspired Guardian AI - a real Google ADK multi-agent system that saves lives in seconds, not minutes.

What it does

Guardian AI coordinates three specialized Google ADK agents: Detection Agent (ingests sensor data), Analysis Agent (applies intelligent risk assessment), and Alert Agent (generates evacuation orders). The system delivers <2 second response time from sensor input to evacuation alert with intelligent Low/Medium/High risk escalation. It's live-deployed on Google Cloud Run with auto-scaling and production-grade monitoring.

How we built it

We built genuine Google ADK agents inheriting from BaseAgent with real session coordination - not mock objects. The architecture includes React+TypeScript frontend, Express.js+FastAPI backend, Python ADK agents, and Docker+Cloud Run deployment. Our intelligent risk engine applies temperature >50°C or smoke >70% = High Risk logic. Production deployment features auto-scaling, BigQuery logging, and IAM security.

Challenges we ran into

Moving from mock to real Google ADK integration required discovering proper Session parameters (id, app_name, user_id). Ensuring multi-agent coordination meant implementing shared session management. Docker deployment issues required fixing incorrect CMD configuration. API synchronization challenges arose from version mismatches between local and deployed code.

Accomplishments that we're proud of

Achieved genuine Google ADK integration with real BaseAgent inheritance and session coordination. Demonstrated sub-2 second emergency response with Detection→Analysis→Alert pipeline. Built production-grade infrastructure with live deployment handling real traffic. Created comprehensive verification system proving agent coordination. Generated life-saving evacuation alerts: "EVACUATE: Server Room - Critical temp: 78°C, Dangerous smoke: 90%".

What we learned

Mastered Google ADK architecture including session management and BaseAgent implementation. Learned multi-agent coordination principles with shared state management and pipeline orchestration. Gained production deployment expertise with Docker containerization and Google Cloud Run. Developed emergency response domain knowledge including risk assessment algorithms and alert escalation patterns.

What's next for Guardian AI

Expanding sensor integration with IoT networks and computer vision. Enhancing AI capabilities with machine learning for predictive analysis. Enterprise deployment with multi-tenant architecture for hospitals and schools. Advanced Google ADK features including real-time streaming agents. Mobile and edge computing for offline capabilities. Global impact initiative with open source release and emergency services partnerships.

Built With

• asyncio
• cors
• css3
• docker
• eslint
• express.js
• fastapi
• git
• github
• google-adk
• google-bigquery
• google-cloud-run
• google-container-registry
• html5
• node.js
• pydantic
• python-3.12
• react-18
• rest-apis
• tailwind-css
• typescript
• uvicorn
• vite