GuardianAI — Real-Time Disaster Safety Agent

GuardianAI is an AI-powered disaster safety assistant designed to help people during dangerous situations such as fires, floods, debris hazards, or collapsed structures.

The system combines computer vision, AI agents, real-time hazard analysis, evacuation planning, and cloud deployment to provide immediate safety guidance.

GuardianAI demonstrates how agentic AI systems can be used for real-world public safety applications.

---

Project Architecture

GuardianAI follows a Composite Multi-Agent Architecture where a coordinator agent delegates tasks to specialized agents.

---

Key Features

1️⃣ Live Hazard Detection (Camera)

GuardianAI can analyze live camera frames and detect hazards such as:

• Fire
• Flood water
• Debris
• Collapsed buildings
• Blocked roads

The AI explains the risk level and recommended safety action.

Example outputs:

---

2️⃣ Disaster Risk Analysis

The system checks disaster alerts and evaluates environmental risks.

Examples:

• Heat advisory
• Flood warnings
• Severe weather alerts

---

3️⃣ Evacuation Guidance

If danger is detected, GuardianAI can:

• Locate nearby emergency shelters
• Provide the exact shelter address
• Generate a Google Maps navigation link

---

4️⃣ Voice Assistance

GuardianAI can generate spoken responses using:

Google Cloud Text-to-Speech

This allows the system to provide hands‑free safety instructions.

---

5️⃣ Emergency Guidance

GuardianAI can help users with safety actions such as:

• Evacuating the area
• Moving toward the nearest shelter
• Following safety instructions during disasters

---

Agent Architecture

GuardianAI implements a Composite Multi-Agent Design Pattern.

This pattern allows a Coordinator Agent to delegate tasks to specialized agents.

Root Coordinator Agent

Handles the main user request and decides which agent should handle it.

Vision Hazard Agent

Analyzes images and detects hazards from camera input.

Disaster Analysis Agent

Evaluates environmental risks such as weather alerts.

Evacuation Planning Agent

Finds shelters and provides evacuation guidance.

---

Agent Flow

User Input
   │
   ▼
Guardian Coordinator Agent
   │
   ├── Vision Hazard Agent
   │       (image hazard detection)
   │
   ├── Disaster Analysis Agent
   │       (weather risk analysis)
   │
   └── Evacuation Planning Agent
           (shelter + maps navigation)

This architecture makes the system:

• Modular
• Scalable
• Easy to extend

---

Tech Stack

AI / Agents

• Google Gemini API
• Google Agent Development Kit (ADK)

Computer Vision

• OpenCV
• Pillow

Backend

• Python

APIs

• Google Maps API
• Google Places API
• Google Geocoding API

Speech

• Google Cloud Text‑to‑Speech

Deployment

• Google Cloud Run

---

Deployment

GuardianAI is deployed on Google Cloud using containerized infrastructure.

Deployment pipeline:

GitHub Repository
       │
       ▼
Docker Container
       │
       ▼
Google Cloud Run
       │
       ▼
Live AI Agent Service

This allows GuardianAI to run as a scalable cloud‑based AI safety system.

---

Running the Project Locally

Install dependencies:

pip install -r requirements.txt

Run the ADK agent interface:

adk web

Run the camera hazard detection:

python camera/camera_capture.py

Controls:

ENTER → analyze current camera frame
Q → exit camera

---

Future Improvements

Possible extensions:

• Automatic emergency contact alerts
• Mobile app integration
• Real-time disaster feed integration
• Multi-language voice assistance
• Continuous live hazard monitoring

---

Team

Amrutha Junnuri

Master's Student — Software Engineering

---

License

This project is for educational and research purposes.