ResQGuide: Agentic Disaster Response Assistant

AI agent that transforms chaotic disaster data into clear, real-time relief priorities for NGOs and governments.

Comment

Inspiration

Floods in India aren’t just seasonal; they’re catastrophic. Every year, millions are displaced, thousands stranded, and lives put at risk. In those critical hours, information pours in chaotically - tweets screaming for help, WhatsApp forwards, fragmented news reports, and government bulletins trickling in. Satellite flood maps exist but remain underused because no human can process them in real time. And so, while cries for help echo online, relief often arrives late, or worse, in the wrong place.

We were inspired by the question: “What if an AI agent could act as a dispatcher, cutting through the noise and telling NGOs exactly where to send help first?”

That idea became ResQGuide: Agentic Disaster Response Assistant.

What it does

ResQGuide is an AI-powered agent that: 1. Collects real-time data from social media, satellite flood maps, and government bulletins. 2. Filters and classifies SOS messages (food shortage, medical emergency, stranded victims). 3. Cross-checks with flood/risk maps to confirm severity. 4. Calculates a priority score for each location (Priority=Severity×Reports×Flood Risk) 5. Visualizes hotspots on an interactive dashboard (red = urgent, orange = medium, green = safe). 6. Sends automated alerts to NGOs via Telegram/WhatsApp with clear action items.

How we built it

• Data Collection: Simulated with past flood tweets, Sentinel-1 flood maps, and government

disaster feeds. • Preprocessing & Geo-tagging: Used NLP to extract location entities (e.g., “Darbhanga, Bihar”), cleaned duplicates, and geocoded coordinates. • Agentic Core: • Agent 1: Detect SOS vs irrelevant noise. • Agent 2: Summarize posts and classify severity. • Agent 3: Cross-verify with satellite flood data. • Prioritization Engine: Implemented a weighted scoring algorithm. • Visualization: Built a dashboard with Streamlit + Leaflet.js. • Notification Layer: Integrated with Telegram Bot API for alerts.

Challenges we ran into

• Messy disaster data: Posts often lacked precise location info.
• Multilingual barriers: Many SOS posts were in Hindi, Bengali, Assamese, etc.
• False positives: Needed to filter out unrelated or duplicate content.
• Time constraints: We simulated a disaster scenario with past data instead of live real-time feeds.

Accomplishments that we're proud of

• Designed a full agentic workflow for disaster response — from raw data → actionable insight.
• Built a working dashboard prototype to visualize hotspots and alerts.
• Managed to integrate both social data + satellite maps into a single prioritization system.
• Created something with potential for real-world humanitarian impact.

What we learned

• How agentic AI frameworks (IBM Granite + ADK) can orchestrate multiple subtasks.
• Basics of handling satellite flood datasets and combining them with text data.
• Practical challenges in multilingual NLP for Indian languages.
• The importance of designing AI with scalability and inclusivity in mind.

What's next for ResQGuide: Agentic Disaster Response Assistant

• Integrate real-time APIs (Twitter/X, ISRO, IMD) for live disaster monitoring.
• Add voice interface for rural accessibility (IVR/WhatsApp voice).
• Deploy as a mobile-first app for NGOs and local governments.
• Extend to other disasters: cyclones, earthquakes, wildfires.
• Partner with NDMA and humanitarian NGOs for pilot deployment.

Built With

Share this project:

Updates

Submission history