RepairVision is an AI-assisted platform for community repair cafés: free, volunteer-run sessions where people bring broken things and fix them together. It runs the café day to day and adds AI where it genuinely helps: working out what is wrong, and deciding whether a repair is worth it.

The problem

• Running a session is chaotic: paper sign-in sheets, a queue of visitors, tracking who is fixing what.
• Diagnosis depends on which volunteers turn up.
• People don’t know if a repair is worth it, so they buy new and the old one goes in the bin.

What it does

• AI Diagnosis (Gemma 4): describe a device and add a photo. Gemma suggests up to three likely faults with supporting and missing evidence, safe checks, and one guided question at a time. Every result is labelled provisional and passes a safety screen that doesn’t depend on the model.
• Repair knowledge base: 100+ team-written entries, the café’s past repairs and iFixit guides back up each suggested fault.
• Repair or Replace advisor: compares the cost of repairing with replacing, allows for a failed repair, and shows the CO₂ and waste a repair avoids, with the working shown.
• Site assistant: a 3D chat character that answers questions from the café’s live data (“What events are on right now?”) and opens the right page. It only sees public data, so it can’t leak anyone’s details.
• Repair café hub: QR check-in, a live repair queue, a waiting-room display, session planning and impact statistics.

How it works

A single Cloudflare Worker serves a SvelteKit app and the API, with D1 (SQLite) for data and R2 for photos. Gemma 4 runs through the Gemini API. Every AI reply must pass a schema and code-level safety checks, and the AI never invents an answer or a price. 199 automated tests run offline with fake AI responses.

Team The Blacklisted: Kanishk Rungta, Keertan Vasani, Adiseshan Ramanan, Priyansh Narang