Inspiration
MatchBack was inspired by how inefficient and unreliable traditional lost-and-found systems are. Most rely on vague descriptions, scattered posts, or manual searching, which often leads to missed connections or false claims. As a team, we noticed how frequently valuable items are lost on campuses and in public spaces, and we wanted to build a system that could intelligently connect the right people while prioritizing safety and trust.
What it does
MatchBack is a smart lost-and-found platform that uses AI to match lost and found items based on location, time, and detailed descriptions. It ranks potential matches by those factors, reducing false claims. It guides both the person who lost an item and the person who found it through a secure, step-by-step claim and return process.
How we built it
We built MatchBack as a cross-platform mobile app using Expo (React Native) for iOS and Android. The backend is a Node.js and Express REST API. It uses a PostgreSQL database managed through Prisma ORM. AI-powered matching is the core of the app. We use a weighted scoring system that evaluates proximity, time difference, and the similarities between the descriptions using the Gemini API. Authentication is handled with JWT. MatchBack includes real-time notifications, match claims, and provides a structured return workflow.
Challenges we ran into
One of our biggest challenges was managing the different user state transitions. We had to ensure both the finder and the owner always see the correct screen based on the current status of a claim. We also had to carefully tune the AI matching logic to balance accuracy and safety, preventing unrelated items from scoring highly while still recognizing genuine matches.
Accomplishments that we're proud of
We’re proud of building a fully functional end-to-end system that combines AI, geospatial logic, and real-world workflows. Successfully implementing a secure claim, approval, and return flow that works for both users while minimizing false claims is also a major accomplishment. We’re also proud of the clean, consistent user experience across the app.
What we learned
Through this project, we learned how to design and implement scalable full-stack applications with real-world constraints. We gained experience integrating AI APIs, designing weighted scoring systems, and managing asynchronous user interactions.
What's next for MatchBack
Next, we plan to improve the AI matching logic by only surfacing matches with confidence scores above 50%, while further refining how the model interprets descriptions so it better understands human phrasing and differently worded but similar descriptions. We also aim to expand support for institutional lost-and-found integrations and add moderation tools to further prevent misuse.
Log in or sign up for Devpost to join the conversation.