Real-time disaster monitoring -- Autonomous drone scouting -- Intelligent rescue strategy generation
ResQNet is an end-to-end platform that combines AI reasoning, drone swarm coordination, and a real-time operator dashboard to accelerate search-and-rescue operations during natural disasters.
The system monitors social media for emerging disasters, deploys autonomous drone swarms to survey affected areas, and generates actionable rescue strategies -- all in real time.
graph TD
A[Monitor - Hourly] -->|Disaster Detected| B{Operator Decision}
B -->|Select Area| C[Scout - Phase 1]
C -->|Send Corners| D[Drone Controller]
D -->|Drone Frames + Logs| C
C -->|Split Zones| E[Zone Grid]
E --> F[Danger Rating]
E --> G[Density Map]
E --> H[Priority Rating]
F --> I[Strategy + Escape Routes]
G --> I
H --> I
I -->|Phase 2 Start| J[Search - Continuous]
I -->|Phase 2 Start| K[Rescue]
J -->|Medical/Threat Alarms| L[WebApp Dashboard]
K -->|Safety Instructions| L
K -->|Operator Suggestions| L
| Module | Description | Port |
|---|---|---|
| Reasoning Agent | AI core -- 4-phase pipeline with LLM/VLM reasoning, detection, heatmaps, and rescue strategy generation | 8000-8003, 8080 |
| Drone Controller | Unity-based drone swarm coordination, flight path management, live feed streaming, telemetry logging | 5000 |
| WebApp | Operator dashboard with maps, heatmaps, drone feeds, and rescue plans | 3000 |
- Scrape X (Twitter) trending topics for potential disasters using twikit
- LLM-based classification and validation of disaster signals
- Deep-dive hashtag analysis for incident details (location, severity, type)
- Persistent reasoning context across monitoring sessions
- Dispatch drone swarm to selected geo-zone via Drone Controller API
- Analyze drone frames with VLM to detect buildings, people, fire, smoke, flood
- Split scouting zone into NxN grid, map drone frames to zones
- Generate per-zone danger rating, people density estimation, and rescue priority
- Produce full rescue strategy with phases and resource allocation
- Generate escape routes and safety zone recommendations
- Continuously process new drone frames for updated detections
- VLM-based medical emergency detection with automatic alarm triggers
- VLM-based threat/risk detection requiring operator attention
- Active alarm queue for real-time operator notifications
- VLM describes scene, LLM generates safety instructions for persons on ground
- VLM tactical assessment, LLM generates operator action plan and resource requests
- Single chat endpoint orchestrates all 4 phases via LLM tool-calling
- 21 tools available across Monitor, Scout, Search, Rescue, and Drone Controller
- Agentic loop with up to 10 chained tool calls per conversation turn
- Maintains conversation history for multi-turn reasoning
| Layer | Technologies |
|---|---|
| AI Models | OpenRouter (LLM + VLM), configurable model selection |
| Backend | Python 3.11+, FastAPI, Uvicorn, httpx |
| Drone Sim | Unity (C#), PID controllers, APF navigation |
| Social Intel | twikit (X/Twitter scraping without API key) |
| Infrastructure | Pydantic v2, Loguru, python-dotenv |
git clone https://github.com/MokhtarOuardi/ResQNet.git
cd ResQNet
# Set up Python environment
cd "Resoning Agent"
pip install -r requirements.txt
# Configure environment
# Create .env with your API keys
# Start all services
python monitor.py # port 8000
python scout.py # port 8001
python search.py # port 8002
python rescue.py # port 8003
python ResQnet_agent.py # port 8080 (unified agent)
# Run the full flow test
python test_flow_simulation.pySee individual module READMEs for detailed setup and configuration instructions.
Developed by Mokhtar Ouardi, Adam Aburaya and Anas Aburaya for the vhack Hackathon.
© 2026 ResQnet Team. All rights reserved. 5