https://3129a065d7e008028e.gradio.live/
About the project
Lost & Found AI for a Smart Campus
The core idea is that people report lost or found items with a photo and a short description, and the AI automatically matches them, so nobody has to scroll through a notice board.
Core features
- Report lost item: photo (optional), description, location, time.
- Report found item: photo, where it was found, where it’s currently kept (security desk, department office).
- AI matching: every new report is compared with the opposite type, and the top matches are shown with a similarity score.
- Notifications: the owner gets an alert when a likely match appears.
- Claim verification: the owner answers a question only the real owner would know (e.g., “what’s the wallpaper?”) before contact details are shared.
- Admin dashboard: for security staff to manage items, see hotspots, and close cases.
How the AI works
| Task | Approach |
|---|---|
| Image matching | Use CLIP (or a similar model) to turn photos into embeddings, then compare with cosine similarity. |
| Text matching | Use sentence-transformers (e.g., all-MiniLM-L6-v2) to embed descriptions like “black Boat earbuds in blue case”. |
| Auto-tagging | Use YOLO or CLIP zero-shot to detect the category (bottle, ID card, phone, bag) and color from the photo. |
| ID card / book OCR | Use EasyOCR or Tesseract to read names or roll numbers, so the system can notify the owner directly. |
| Final score | A weighted mix: 0.4 × image + 0.3 × text + 0.15 × location proximity + 0.15 × time proximity. |
The location and time weighting is what makes it “smart campus” rather than just a generic image search. An item lost in the library yesterday is far more likely to match one found in the library than one found at the hostel.
Suggested architecture
- Frontend: React web app, or Flutter if you want a mobile app. Students will mostly use phones.
- Backend: FastAPI (Python), since your ML code lives in Python too.
-
Database: PostgreSQL with
pgvectorto store and search embeddings. FAISS also works. - Storage: Cloudinary or local storage for images.
- Notifications: email, Firebase push, or Telegram bot.
- Campus map: a list of named zones (library, canteen, labs, hostels) as location options.
Build order
- Week 1: Basic app with report forms, database, and a listing page.
- Week 2: Text matching, which is easy and already works well.
- Week 3: Image embeddings and combined scoring.
- Week 4: Notifications, claim verification, and admin dashboard.
- Final: OCR, a hotspot heatmap, and your demo and report.
Ideas to stand out
- Heatmap of where items are most often lost.
- QR tags that students stick on their belongings, so scanning one reaches the owner directly.
- A chatbot (“I lost my grey water bottle near the canteen”) that files the report conversationally.
- Privacy handling, such as blurring faces or personal details on ID card photos.
I can go deeper on any part: the database schema, the matching code, a project report outline, or a working prototype. To point you in the right direction, what’s your deadline and team size, and is this for a college submission or a real deployment?
Built with
openaigemma
AI tools
Hand-coded, no AI
Prize Categories Entered
- Submitted
Best Use of Gemma 4