FindLoop

Inspiration

Losing an important item is stressful, and most lost-and-found systems depend on exact keywords and manual searching. We wanted to create a simpler solution that understands both images and normal descriptions, searches within the relevant location, and connects finders with owners. We also wanted FindLoop to support older adults and people with reduced vision. This inspired Look & Ask, which helps users locate nearby objects using a photograph and a voice question.

What it does

FindLoop allows users to: Report lost or found items using a photo or text description. Use GenAi to extract the item’s category, colour, brand, and identifying details. Select a location using GPS, Google Maps, or text search. Search for nearby matches using item details, geohashing, and distance. Review and manually select likely matches. Send messages and item locations through a persistent Firestore inbox. Search again after saving a lost or found report. Display the actual uploaded item photograph. Mark successfully returned items as reunited. Use Look & Ask to photograph an area, ask a question by voice, and hear where an object is located.

How we built it

We built FindLoop using Flutter and Dart for the mobile application. GenAi analyzes item photographs, text descriptions, and Look & Ask scenes. Google Maps, Places search, and device location services support location selection. Firebase Authentication provides Gmail login. Cloud Firestore stores reports and inbox messages, while Firebase Storage stores uploaded photographs. Authenticated Cloud Functions validate reports and securely create messages in the correct user’s inbox. Matching combines structured AI-generated item details with geohash-based location filtering and Haversine distance calculations. We used OpenAI’s GPT-5.6-sol through Codex as our development collaborator. It helped us design the architecture, implement features, debug device and Firebase issues, improve the user experience, write automated tests, and prepare the production release.

Challenges we ran into

One major challenge was creating separate and correct lost and found workflows. A found report needed to search existing lost reports instead of creating another lost item. Location handling was also challenging because we needed to support GPS, map selection, and text search while restricting matches to the relevant area. We initially used FCM for notifications, but device-token management made the flow unreliable. We replaced it with durable Firestore inbox messaging, allowing messages to remain available when users reopen the application. Other challenges included securely validating matches, storing actual uploaded photographs, processing description-only reports, managing Firebase permissions, and designing Look & Ask for users with reduced vision.

Accomplishments that we're proud of

We are proud that FindLoop provides a complete end-to-end lost-and-found experience. The application supports both photos and normal descriptions, performs location-aware matching, requires human confirmation before contacting an owner, and stores messages securely. We are especially proud of Look & Ask because it expands FindLoop beyond lost-item reporting and provides practical assistance to people who may struggle to see small objects around them. We also created automated tests, secure Firestore and Storage rules, authenticated backend validation, and a working Android release tested on a physical device.

What we learned

We learned that successful AI applications need both intelligent analysis and clear user control. AI can rank likely matches, but users should make the final decision. We also learned that location is as important as visual similarity for lost-and-found matching. Combining AI-generated details with distance filtering produces more useful results. Finally, we learned the importance of designing reliable communication. Persisting messages in Firestore created a simpler and more dependable experience than relying on device notification tokens.

What's next for FindLoop

Next, we plan to: Add multilingual voice and text support. Improve matching with semantic embeddings and user feedback. Support schools, offices, transport stations, and residential communities. Add optional verified handover workflows. Expand Look & Ask with more accessible guidance. Improve privacy controls for sensitive locations. Release FindLoop on both Android and iOS. Build moderation and duplicate-report detection for larger communities.

Built With

• cloud-firestore
• cloud-functions
• codex
• dart
• firebase-authentication
• firebase-storage
• flutter
• geolocator
• google-maps
• google-places
• openai-gpt-5.6-sol
• speech-to-text
• text-to-speech