Biomechanical ML coaching that perfects your form, and tracks your workout in real-time.

Built at BACSA Hacks 2026

Next.js · TailwindCSS · Three.js · MediaPipe · Flask · Gemini · MongoDB

70% of gym injuries are caused by bad form. People new to working out are 2.5× more likely to get injured and a personal trainer costs hundreds of dollars an hour. We built FormCop, a browser-based ML fitness coach that watches you work out in real time, counts your reps, corrects your posture, and stores your full workout history.

• 3D animated exercise selector with Mixamo FBX models rendered via Three.js / R3F
• Real-time pose tracking using a custom-tuned MediaPipe ML model with EMA smoothing
• Live coaching overlay via instant text instructions while you move
• Automatic rep counting with per-rep joint angle logging (Pushup, Squat, Bicep Curl)
• Duration tracking for timed exercises (Plank, Wall Sit, Handstand)
• AI form analysis powered by Gemini 2.5 Flash Lite — full set feedback after every set
• Workout session tracker to see every set and rep logged in the current session
• Full workout history dashboard with all past sessions stored in MongoDB

• Node.js 18+
• Python 3.10+
• MongoDB Atlas account (free tier works)
• Google Gemini API key

git clone https://github.com/BryanYeeee/FormCop.git
cd formcop

cd frontend
npm install
npm run dev

The Next.js dev server starts at http://localhost:3000

cd backend
python -m venv venv && source venv/bin/activate   # Windows: venv\Scripts\activate
pip install -r requirements.txt

Create a .env file in the backend folder:

GEMINI_API_KEY=your_gemini_api_key_here

Update the MongoDB URI in mongo/mongo.py with your Atlas connection string, then:

python app.py

The Flask API starts at http://localhost:8080

In your Atlas cluster create a database named bacsa with a collection named userinfo. Insert at least one user document:

{ "username": "yourname", "password": "yourpassword", "workouts": [] }

1. User logs in → POST /mongo/ → username stored in AuthContext + localStorage
2. User selects exercise → ExerciseSelector → router.push('/workout/[id]')
3. PoseTracker runs MediaPipe on webcam frames, calculates joint angles via processPose
4. Each rep is stored in repTracking state with peak angles + timestamp
5. User taps Submit Set → Gemini analysis → FeedbackPanel displayed
6. User taps Finish → addSet() called on WorkoutContext → router.push('/tracker')
7. User either repeats steps 2-6 or taps End Workout → endWorkout(username) → POST /mongo/new → clears context → router.push('/home')
8. Dashboard fetches GET /mongo/get?username=... → renders collapsible workout cards

• Rian Mehta (BCS)
• Faiyad Ahmed Masnoon (BCS)
• Bryan Yee (BCS)
• Martin Lam (KPE)

Built in under 12 hours at BACSA hacks