NeighborNet AI

An AI-powered platform that rescues surplus food, connects communities with volunteers, and optimizes resource distribution during everyday operations and disaster emergencies.

Comment

Inspiration

Every day, restaurants, bakeries, and supermarkets have surplus food that could support people in need, yet much of it goes to waste due to inefficient coordination. During disasters, the same communities face additional challenges in distributing essential supplies quickly and fairly.

We were inspired to build a solution that works in both situations: reducing food waste during normal operations and strengthening community resilience during emergencies.

This led us to create NeighborNet AI — an intelligent food rescue and disaster resource coordination platform.

What it does

NeighborNet AI is a dual-mode platform connecting food providers, NGOs, volunteers, and community coordinators.

Normal Community Mode

  • Allows businesses to register surplus food.
  • Connects available resources with verified NGOs and community organizations.
  • Recommends suitable allocations based on demand, location, availability, and food safety.
  • Coordinates volunteer pickups and deliveries.
  • Tracks resource distribution and community impact.

Disaster Response Mode

  • Enables authorized coordinators to activate disaster events.
  • Helps shelters and NGOs request urgent food, water, and essential supplies.
  • Prioritizes resource allocation based on verified needs.
  • Recommends alternative assignments when volunteers become unavailable or deliveries are disrupted.

The platform includes a smart unified dashboard, multi-role access, geographic visualization, explainable recommendations, and human-approved allocation workflows.

How we built it

We developed NeighborNet AI as a full-stack web application using modern open-source technologies.

Frontend: React, Vite, TypeScript, Tailwind CSS, and Mantine UI for a responsive, role-aware user experience.

Backend: Python, Django, and Django REST Framework for secure APIs, authentication, permissions, and resource management.

Database: PostgreSQL with PostGIS for resource records, geographic information, and location-based matching.

Optimization: Google OR-Tools CP-SAT for constraint-based resource allocation, supported by first-come-first-served and nearest-compatible matching baselines.

Data Analysis: Pandas and NumPy for processing and evaluating resource distribution performance.

Maps and Analytics: MapLibre GL JS, OpenStreetMap, and Apache ECharts for geographical visualization and interactive analytics.

We also incorporated publicly available Sri Lankan geographic, historical weather, disaster-report, and food-waste research data to provide real-world context.

For development and reproducible algorithm testing, we used clearly labeled synthetic food donation, recipient-demand, and volunteer scenarios.

Challenges we ran into

One major challenge was designing a system that supports everyday food redistribution and disaster relief without duplicating workflows.

We also had to address multi-role authentication, ensuring that someone could be both a food provider and a volunteer using a single account.

Developing a reliable optimization engine was another challenge. Recommendations needed to respect resource availability, food safety, recipient requirements, deadlines, geographical constraints, and volunteer capacity.

We encountered inconsistencies in early benchmark calculations and addressed them by introducing clearer metric definitions, reproducible evaluation datasets, and additional validation tests.

Access to real local food-donation records was limited, so we separated genuine public datasets from synthetic operational scenarios rather than presenting simulated activity as real-world outcomes.

Accomplishments that we're proud of

  • Built a functional full-stack platform supporting both normal and disaster operations.
  • Implemented unified authentication and multiple approved roles per account.
  • Developed an intelligent allocation engine using Google OR-Tools.
  • Integrated resource matching, volunteer assignments, and disruption recovery.
  • Created explainable recommendations with coordinator approval controls.
  • Integrated PostgreSQL/PostGIS for geographic resource management.
  • Developed reproducible comparisons between optimization and baseline matching algorithms.
  • Incorporated Sri Lankan public data sources to support our pilot context.
  • Passed 39 reported backend tests and successfully built the React frontend.

In our seven synthetic evaluation scenarios, the OR-Tools engine achieved an average reported demand fulfillment rate of 88.9%, compared with approximately 50.3% for first-come-first-served matching. These are controlled simulation results, not verified real-world food rescue outcomes.

What we learned

Building NeighborNet AI taught us that effective AI is not simply about using complex models. It is about making reliable, explainable, and useful decisions based on available data.

We gained practical experience in full-stack development, geospatial databases, optimization algorithms, role-based access control, API integration, and evaluation methodology.

We also learned the importance of maintaining data integrity, comparing intelligent approaches against simple baselines, and keeping humans involved in critical decisions.

Most importantly, we discovered how technology can connect sustainability, community collaboration, and disaster resilience within a single platform.

What's next for NeighborNet AI

Our next goal is to move from a working prototype toward real community deployment.

We plan to:

  • Collaborate with local restaurants, bakeries, NGOs, and volunteer groups in Sri Lanka.
  • Validate the platform using consent-based, real operational data.
  • Improve route planning with reliable road-network information.
  • Introduce surplus food and resource-demand forecasting when sufficient historical data becomes available.
  • Expand disaster response coordination with more robust disruption handling.
  • Strengthen accessibility, mobile usability, and deployment reliability.
  • Scale gradually from the Moratuwa–Colombo South pilot region to additional communities.

Our long-term vision is to make NeighborNet AI a practical platform where surplus resources reach people who need them, and communities can respond more effectively when circumstances change.

Built With

Share this project:

Updates