RePair

App Home
Tech Stack
Safety
Toolz
Repair Sample

Inspiration

All it takes is one slip-up. You’re lying in bed at 2 AM, scrolling for “just one more reel,” and then bam: your phone drops on your face, hits the floor, and the screen suddenly looks like abstract art.

And it’s not just phones. Between the four of us, we’ve cracked laptops, fried controllers, and broken more kitchen appliances than we’d like to admit. The problem is universal: everything breaks, and fixing anything is way too hard.

Repair guides are scattered across old forums, hour-long YouTube videos, and blurry PDFs that look like they were scanned in a garage. Manufacturers hide screws, glue parts together, and make repair documentation impossible to find. Right-to-Repair laws are finally giving people the legal right to fix their devices, but no one has built a tool that makes that right actually usable.

So we built RePair: a simple, fast way to understand what broke and how to fix it, while encouraging reuse and reducing e-waste. A tool that turns repair from a frustrating guessing game into something visual, intuitive, and sustainable.

What it does

RePair turns a single photo of a broken device into a clear, visual repair guide. Just upload an image, and RePair instantly generates a 3D exploded view of the object, identifies its components, and lays out step-by-step repair instructions. You can even download an IKEA-style instruction PDF for fixing it later, offline.

We built RePair to be simple: no searching, no forums, no 47-minute YouTube videos—just take a picture and start fixing. Whether it's a cracked phone screen, a loose controller joystick, or a laptop that suddenly stops behaving, RePair gives users a beginner-friendly, visual explanation of exactly what’s inside and how to repair it.

How we built it

We built RePair using a hybrid AI + API + 3D approach:

Frontend: Next.js + TypeScript for a fast, clean UI
Backend: Python FastAPI for routing images, orchestrating AI calls, and generating structured outputs
AI Reasoning: Gemini for interpreting images, generating steps, and enriching instructions
Repair Data: iFixit API for verified teardown guides
3D Visualization: Three.js for animating exploded diagrams
PDF Generation: Automated pipeline that formats the instructions into printable repair sheets

The backend orchestrator acts as the “brain,” combining reasoning, device detection, 3D diagram generation, and instructions into one cohesive flow.

Challenges We Ran Into

One of our biggest challenges was trying to get SAM3 up and running. We seriously underestimated how difficult it would be to install, configure, and run such a heavy vision model in a hackathon environment. Between environment mismatches, CUDA/GPU issues, and dependency conflicts, we spent hours trying to make it work before accepting that we needed to pivot. We still want to integrate SAM3 in the future, but this weekend, it wasn’t happening.

Another major challenge was building a full end-to-end pipeline under extreme time pressure. RePair touches nearly every layer of a modern stack: frontend UX, backend routing, LLM processing, 3D rendering, and generating downloadable repair PDFs. Each component worked individually early on, but stitching them together into one smooth workflow was much harder than expected.

And of course, the biggest constraint was simply time, combined with the overly ambitious idea we were determined to build. We pushed ourselves to the edge, but ultimately we’re proud that we delivered a fully working system… in literally the last hour.

Accomplishments that we're proud of

Despite everything, we shipped a fully functioning system (finished in the last hour of the hackathon). We’re proud that:

The upload → analyze → 3D explode → instruction flow works end-to-end
The PDF generation pipeline is smooth and reliable
The interactive 3D component view feels intuitive and fun

What we learned

We learned one of the most important hackathon lessons: start with a simple prototype.

We spent a lot of time fighting with advanced tooling (like trying to download and configure SAM3), and that sunk almost a whole day. In a hackathon setting, building the reliable baseline first, and saving the risky integrations for later, would have saved us a lot of pain.

We also learned how to stitch together AI reasoning with 3D rendering in a way that feels natural, and how to design a repair experience that’s both technically correct and beginner-friendly.