Disaster Task Sending Strategy:

Small urgency can be managed by volunteers.
Medium-scale urgency* are handled primarily by first responders.
Large-scale complex urgency* require combined efforts of government bodies and volunteers.
This structure ensures volunteer safety while enabling scalable response.

Resource Assumption: All aid resources (e.g., shelters, medication stations) are assumed to be static and not mobile.

Voice Input: Voice input is not accepted for reporting disasters due to:

Environmental noise commonly present in disaster situations can interfere with voice recognition accuracy.
Language or accent barriers may prevent accurate interpretation of spoken reports.
Individuals may be in distress or panic, which can affect speech clarity and coherence.

Image Analysis Strategy:

We do not use VLMs like CLIP for classification due to high computational cost and low control over label specificity.
Instead, a custom CNN model is used to classify disaster types and count people—this is faster and cheaper after training.
Additionally, we leverage Gemini Flash 2 API, a multimodal model,which also a VLM capable model for more robust image understanding.

Weather Data:

We use the OpenWeather API for Map precipitation data filter relevant to disasters.
Although the API allows only 60 calls per minute, this is considered acceptable under expected usage.

AI Workflow Architecture

Workflow 1: Multiagent Emergency Report Analysis

Triggered: When first emergency call/photo is submitted Purpose: Create government and citizen reports, predict future impact, validate request authenticity

2 AI Agents:

Government Analysis AI Agent: Creates official response reports using Gemini AI
Citizen Survival AI Agent: Generates public safety instructions using Gemini AI

3 Data Collection Tools:

Computer Vision Tool: Analyzes emergency photos using CNN + YOLO models
Weather Tool: Gets current conditions (Open-Meteo API)
Disaster History Tool: Fetches past disasters (GDAC API)

5 System Coordinators:

Parallel Data Coordinator: Runs weather + history+ computer vision tools simultaneously
Data Validation Coordinator: Checks data quality and completeness
Parallel AI Coordinator: Runs government + citizen agents simultaneously
Final Review Coordinator: Evaluates all outputs for acceptance/rejection
Master Coordinator: Orchestrates entire multiagent workflow

Multiagent Interaction Flow:

          First Emergency Request
                        ↓
Parallel Data Collection → Weather Tool + History Tool +  Computer Vision Tool (simultaneous)
                        ↓
Data Validation → Quality check by validation coordinator
                        ↓
Parallel AI Analysis → Government Agent + Citizen Agent (simultaneous)
                        ↓
Final Coordination → Master coordinator reviews all agent outputs → Decision

Agent Collaboration: Each AI agent works independently on specialized tasks while coordinators manage data flow, timing, and final decision-making across all agents.

Workflow 2: Task Creation & Dispatc

Categories

Current Category: cnn-classification

TetraNeurons