AEGIS Smart Protocol

AI Agents Powering Trustless Disaster Relief

Overview

AEGIS Smart Protocol is a fully autonomous, multi-agent disaster response system designed to remove human bottlenecks from emergency aid. The protocol utilizes a three-stage pipeline (Detection → Verification → Disbursement) to analyze satellite/drone imagery in real-time and execute trustless cryptocurrency payments to relief organizations immediately upon verifying a disaster event.

The Problem

Traditional disaster relief systems suffer from:

• Slow Response Times: Manual verification and bureaucratic approval processes delay critical aid
• Lack of Transparency: Donors can't track where their money goes
• Human Bottlenecks: Relief organizations wait days or weeks for funding approval
• Fraud & Misallocation: No automated verification of actual disaster events

The Solution: Three AI Agents

1. Watchtower Agent

The eyes of the system. Runs parallel computer vision algorithms to detect disasters from raw aerial imagery:

• Fire Detection: HSV color space analysis for flame signatures
• Flood Detection: Water body expansion analysis
• Structural Damage: Edge detection for collapsed buildings
• Casualty Detection: Object detection for emergency scenarios

All detections require >60% confidence thresholds to proceed.

2. Auditor Agent

The brain of the system. Verifies events by:

• Cross-referencing multiple data sources
• Calculating "Human Impact Scores" based on population density
• Preventing false positives and fraud attempts
• Generating transparent audit trails

3. Treasurer Agent

The hands of the system. Autonomously manages an Ethereum wallet to:

• Sign and broadcast transactions to relief organizations
• Handle gas fee estimation and nonce management
• Execute payments without human intervention
• Maintain transaction records on-chain

Technical Architecture

Core Technologies

• Python 3.9+: Agent orchestration and logic
• OpenCV: Computer vision for disaster detection
• NumPy: Data analysis and scoring algorithms
• Redis: Message queuing between agents

Blockchain Layer

• Ethereum (Sepolia Testnet): Smart contract deployment
• Web3.py: Blockchain interaction
• Eth-account: Wallet management
• Infura RPC: Node connectivity

Infrastructure

• Docker: Containerized deployment
• Flask: REST API for frontend communication
• Heroku / AWS: Cloud hosting

Frontend

• Vanilla HTML/CSS/JS: Lightweight interface
• Neobrutalist UI Design: Bold, transparent design philosophy

Key Features

Real-time Detection

Processes satellite and drone imagery in real-time using parallel computer vision algorithms across four disaster categories.

Autonomous Verification

Multi-layered verification system that calculates human impact scores and cross-references data to prevent false positives.

Trustless Disbursement

Blockchain-based payment system that executes cryptocurrency transfers to verified relief organizations without human intervention.

Glass Box Transparency

Every decision made by the AI agents is logged and auditable, making the autonomous system transparent to donors and stakeholders.

Technical Challenges & Solutions

Challenge 1: Agent Orchestration

Problem: Coordinating three independent AI agents asynchronously. Solution: Implemented Redis message queuing to enable reliable inter-agent communication with retry logic.

Challenge 2: Fire Detection Accuracy

Problem: Distinguishing actual fire signatures from similar visual patterns (sunsets, red objects). Solution: Developed robust HSV color space analysis with multiple threshold checks and context-aware filtering.

Challenge 3: The Oracle Problem

Problem: Bridging off-chain computer vision data with on-chain financial execution. Solution: Created a trusted Auditor agent that acts as a verified oracle, cryptographically signing disaster confirmations.

Challenge 4: Gas Management

Problem: Managing testnet ETH liquidity and gas estimation during high-load scenarios. Solution: Implemented dynamic gas estimation with fallback mechanisms and transaction queue management.

Impact & Learnings

What I Learned

• DeFi Integration: Deep integration of Web3.py protocols with real-world physical data from OpenCV
• Transparent AI: Designing "Glass Box" interfaces that make autonomous AI decision-making auditable
• DevOps: Containerizing complex multi-service architectures (API + Workers + Redis) for consistent deployment
• Computer Vision: Advanced image processing techniques for disaster detection

Event

Built for Calcutta Hacks by team Cloud Chingri

Links

• Devfolio Project
• Demo Video
• Event Page

Status

Currently in Prototype stage with plans for mainnet deployment and integration with established relief organizations.