FormFit AI

AI-powered lifting coach — hackathon build.

Point a camera at your lift (or upload a clip) and the app detects the exercise and camera angle, counts reps, grades your form rep by rep, and reads your effort from rep count, rep-speed decay and facial strain cues — then coaches you live.

Status

Frontend complete, AI analysis currently simulated in the browser. The pose overlay, exercise classification, rep timing, form scores and effort metrics are produced by a demo engine (src/lib/simulation.ts) that mirrors the shape of the real model output, so the backend can be swapped in without UI changes.

Stack

• React 19 + TypeScript + Vite
• Tailwind CSS + shadcn/ui
• Recharts (rep tempo / form charting)
• lucide-react icons

Run it

npm install
npm run dev

Then open http://localhost:3000

App flow

• / — landing page
• /session — live set analysis:
  • Start from camera (getUserMedia), video upload, or demo mode
  • Movement classification (exercise + camera angle, with confidence)
  • Animated pose-estimation overlay with key-joint angle readout
  • Rep counter, rep tempo, live form score (0–100) with coaching cues
  • Effort meter (0–100) fusing reps, tempo decay and strain cues
  • End-of-set summary with stats and a coach's note

Where the AI plugs in

src/lib/simulation.ts exports simulateRep(), ExerciseDef and the feed/effort types consumed by src/pages/Session.tsx. Replace the simulated rep generator and classifier with real inference (pose model + form/effort heuristics) and the UI will render it unchanged.