Lost Link
About the project
LostLink AI helps students, faculty and staff reunite with lost belongings. People report lost or found items, and the app uses Gemma 4 to compare the reports and suggest possible matches. It explains what is similar and what differs, so owners and finders don’t have to rely on identical wording.
Full description:
Students regularly lose IDs, wallets, keys, bottles, headphones and notebooks. Lost-and-found reports are written in inconsistent words (“black wallet” vs “dark purse”), and manual announcements rarely connect the owner with the finder.
LostLink AI is a simple web app that fixes this. A student reports a lost item and a finder submits a found-item report, optionally with a photo. Python rules first filter out clearly unrelated candidates. Gemma 4 (gemma-4-26b-a4b-it, through Google GenAI) then compares the remaining item details and returns:
an estimated similarity score
matching characteristics
conflicting characteristics
a short explanation
a recommended next step
If Gemma 4 is unavailable, the app falls back to basic keyword matching, clearly labelled as a fallback.
Key features
Report forms for lost and found items.
Browse and filter by status, category, location and keywords.
A dashboard with totals, recent reports and category stats.
Optional SMS alerts through Twilio. A lost-item reporter must give an international-format phone number and explicitly opt in. Likely matches scoring 70 or more can trigger an alert saying the item “may match”.
Responsible AI and privacy
A match is a lead for human review, not proof of ownership.
The score is an AI-estimated similarity, not a calibrated probability.
The AI is instructed not to invent details and to note contradictions.
Contact details are not shown publicly in full.
Claimants verify a private detail, and valuables and IDs need manual staff verification.
Tech stack: Python, Streamlit, SQLite, Gemma 4 via Google GenAI, Twilio (optional) and Pytest.
Built with
AI tools
Prize Categories Entered
- Submitted
Best Use of Gemma 4
- Submitted

Best Open-Source AI Project