WasteFellow

Inspiration

Walking through UCSC dining halls, we noticed mountains of uneaten food being thrown away daily. Students take more than they can eat, dining staff over-prepare popular items, and there's no data to guide better decisions. We realized that what gets measured gets managed – so we set out to build an AI-powered system that turns food waste from an invisible problem into actionable insights.

What it does

WasteFellow is an AI-powered food waste monitoring system for UCSC dining halls. Cameras on conveyor belts automatically scan returned plates, using computer vision to identify and measure leftover food in real-time. The dashboard displays live metrics across all five dining halls—total waste, cost impact, plates scanned, and waste per student—while breaking down data by meal time and food category. The system's AI engine analyzes patterns and generates specific recommendations for dining staff, such as "Reduce Rice & Grains by 35% at lunch" or "Increase Fruits by 10% at dinner." By turning every plate into actionable data, WasteFellow helps dining services optimize portions, cut costs, and eliminate food waste.

How we built it

We built it using a streamlined tech stack optimized for real-time food waste detection and analysis. The frontend runs on Streamlit with Python, providing an intuitive dashboard that updates instantly as new data flows in. We trained a custom object detection model using Roboflow, feeding it hundreds of images of plates with various food items to teach it to recognize and classify leftovers like rice, vegetables, proteins, and fruits. The model connects via Roboflow's API, processing uploaded plate images and returning detected food items with confidence scores. We store all detection data, meal statistics, and dining hall metrics, structuring tables for daily stats, meal-specific waste, dining hall breakdowns, and individual detections. The recommendation engine uses a custom algorithm that calculates waste percentages by meal and food type, then applies smart thresholds—flagging items over 25% waste for reduction and items under 8% waste for potential increase. The system automatically categorizes detected food items, estimates weight based on portion sizes (approximately 0.15 lbs per detected item), updates all relevant database tables, and triggers dashboard refreshes to display real-time insights. Everything is containerized in a single application that can be deployed on any server with camera access, making it ready for immediate testing in actual dining halls.

Challenges we ran into

Our biggest challenge was integrating Roboflow's detection API with Streamlit, syncing database updates across multiple tables, and ensuring instant frontend refreshes required extensive debugging. Every plate scan had to simultaneously update daily stats, meal-specific waste, dining hall data, and food categories without creating bottlenecks. We also struggled with training the AI to distinguish similar foods and estimating weight accurately. While we built a working proof-of-concept, perfecting this data pipeline is essential because we plan to launch WasteFellow as a real project at UCSC dining halls and expand to other college campuses. This hackathon proved it works—now we're focused on making it production-ready to actually tackle food waste at scale.

Accomplishments that we’re proud of

We’re proud of leading an initial idea into a functional product with a clear sustainability mission. Through creating numerous iterations of our project, we validated a real waste-reduction problem and built a solution designed around user needs.

What we learned

We learned that simple, easy-to-use tools are essential for encouraging sustainable behavior. Getting feedback early from mentors and communicating clearly as a team helped us build a better and more effective project.

What’s next for WasteFellow

We plan to implement WasteFellow in UCSC dining halls by installing cameras on return lines and working with dining staff to test the system in real conditions. Using feedback from this rollout, we’ll improve accuracy, usability, and recommendations, with the long-term goal of expanding WasteFellow to other college campuses that want an easy way to track and reduce food waste.