BhookhBuster a.k.a HungerBuster

Inspiration

The inspiration for HungerBuster came from observing two persistent issues on campus: large amounts of food being wasted after events or dining hall operations, and students regularly searching for affordable or free food options. We wanted to create a system that bridges this gap by connecting surplus food with students in real time, reducing waste while improving accessibility.

What it does

HungerBuster collects dining hall data, identifies available food, and uses AI-powered suggestions to help students discover options that match their preferences, dietary needs, or even mood. It streamlines food discovery by combining web scraping, structured data, and large-language-model recommendations. The system also supports workflows for administrators to post or manage available food, creating a centralized platform for both students and food providers.

How we built it

We began by using Claude as a collaborative tool to design the system structure, refactor code, and quickly prototype components. The project is primarily written in Python with Flask handling routing and API endpoints. Early versions included JavaScript components, but we eventually consolidated logic into Python to improve consistency and development speed.

We built a modular architecture consisting of a web scraper for dining hall APIs, a JSON-based caching layer, a Claude API integration service, JWT authentication, and multiple flows for retrieving data, generating suggestions, and managing admin actions. Each piece was designed to operate independently while still working together through well-defined interfaces.

Challenges we ran into

One major challenge was our initial attempt to build the system using TypeScript and Tailwind CSS, which added front-end overhead and slowed early progress. This led us to pivot toward a backend-first approach with Python.

Data inconsistencies were another challenge. The dining API often returned incomplete or irregular fields, which required building a transformation layer to standardize schemas before storing or processing data.

Integrating multiple workflows also required careful coordination to avoid stale cache states or conflicting results. Rate limits from the Claude API forced us to implement batching, fallback logic, and response handling.

JWT-based authentication introduced complexity around token expiration and validation. Additionally, early mixed-language development (Python and JavaScript) created CORS issues and type mismatch problems before everything was unified in Python.

Accomplishments that we're proud of

We are proud of successfully integrating AI-powered recommendations into a system that helps reduce food waste. The modular backend design, the ability to scrape and clean real dining hall data, and the use of JWT for secure workflows all reflect strong structural engineering. The pivot from TypeScript to Python allowed us to move quickly while still maintaining project quality. Overall, we created a functioning, extensible foundation for matching surplus food with students who need it.

What we learned

Through this project, we learned how to integrate the Claude API effectively, design prompt workflows, and incorporate AI into practical applications. We gained experience with web scraping, schema cleaning, and implementing multi-step data pipelines. We built a secure authentication flow using JWT and learned how to coordinate caching, scraping, and AI logic without causing sync issues. More broadly, we learned how to shift technologies mid-project, how to refactor modular code, and how to design systems that can scale beyond a single workflow.

What's next for HungerBuster

The next phase focuses on turning HungerBuster into a scalable, community-wide platform. This includes replacing temporary JSON storage with a robust database (such as PostgreSQL or MongoDB), enabling high-availability API workflows, and improving caching and synchronization accuracy.

We plan to expand beyond campus dining halls by partnering with local eateries in Ithaca, allowing restaurants to post leftover or discounted food that residents and students can access. With proper infrastructure and authentication in place, the platform could also be extended to other university campuses, enabling a network where food surplus and demand are efficiently matched across multiple regions.