ResQPlan

Inspiration

Floods in Chennai and across India show a recurring pattern: resources are available, but they don’t reach people in time. Trucks get stuck, boats are underutilized, and some neighborhoods receive excess while others wait for days. We wanted to solve this gap by building an AI-driven system that makes disaster relief smarter, faster, and fairer.

What it does

Our project, AI Relief Allocator, optimizes the distribution of food, water, and medical supplies during disasters. Takes in zone data (population, severity), depots (stock), and assets (trucks/boats). Runs an optimization engine that balances efficiency, fairness, and speed. Adapts to disruptions (e.g., blocked roads, sudden SOS requests). Provides an interactive map visualization showing before vs. after allocation. Generates plain-English rationale for why resources were allocated in a particular way.

How we built it

Backend: Python with OR-Tools for linear/mixed-integer programming. Distance calculation: Pre-computed Haversine matrix with truck/boat speeds and congestion factors. Frontend: Streamlit dashboard with maps, sliders for efficiency/fairness trade-offs, and scenario simulation. Data: Used open datasets for Chennai wards, relief centres, and flood extent (2015) to create a realistic simulator. Simulation: Added “event triggers” (road block, demand spike) to test re-optimization in real time.

Challenges we ran into

Designing a multi-objective optimizer that balances fairness as well as efficiency. Making the system robust under infeasible cases (when trucks can’t reach certain zones). Building a realistic demo within hackathon constraints — solved by combining real datasets with simulated live events. Keeping the UI simple while conveying complex KPIs like Gini fairness and robustness scores.

Built With

• csv
• fastapi
• folium
• geojson
• javascript
• leaflet.js
• or-tools
• pandas
• pulp
• python
• streamlit
• typescript