🏋️ FormCheck AI

Real-time AI gym coach that uses computer vision to detect exercise form, count reps, and provide live voice coaching — all through your webcam.

Demo

Try it live →

Landing Page →

What It Does

FormCheck AI watches you exercise through your webcam and:

• Detects your pose — Tracks 33 body landmarks in real-time using MediaPipe
• Analyzes your form — Calculates joint angles, checks depth, alignment, and flags errors
• Coaches you with voice — Generates natural coaching cues using an LLM and speaks them aloud
• Counts reps and sets — Automatic tracking with workout history saved per user

Supported Exercises

Exercise	What It Tracks
Squats	Knee angle, back angle, depth status
Push-ups	Elbow angle, body alignment, hip position
Biceps Curls	Elbow angle, shoulder stability, swing detection
Shoulder Press	Elbow angle, arm extension, back arch
Lunges	Front knee angle, torso angle, balance status

Tech Stack

Component	Technology
Pose Detection	MediaPipe Pose Landmarker
Video Processing	OpenCV
Web Interface	Streamlit
Real-time Streaming	streamlit-webrtc + WebRTC
AI Coaching	Groq API + LLaMA 3.3 70B
Voice Synthesis	gTTS (Google Text-to-Speech)
Database	SQLite
Language	Python 3.11

How It Works

1. Webcam feed is captured via WebRTC and processed frame-by-frame
2. MediaPipe extracts 33 pose landmarks from each frame
3. Exercise detectors calculate joint angles and detect rep stages (up/down)
4. Form analysis identifies errors (e.g., knees caving, hips sagging, back arching)
5. Groq LLM generates a short coaching cue based on the detected issue
6. gTTS converts the text to speech and plays it back in real-time
7. SQLite persists workout history per user

Project Structure

├── main.py                          # App entry point
├── requirements.txt                 # Dependencies
├── .streamlit/config.toml           # Theme configuration
├── core/
│   └── base_exercise.py             # Abstract base class for exercises
├── detectors/
│   ├── squat.py                     # Squat detection logic
│   ├── pushup.py                    # Push-up detection logic
│   ├── biceps_curl.py               # Biceps curl detection logic
│   ├── shoulder_press.py            # Shoulder press detection logic
│   └── lunges.py                    # Lunge detection logic
├── ml_models/
│   └── pose_landmarker_full.task    # MediaPipe pose model
├── services/
│   ├── auth/login_wall.py           # User login
│   ├── coaching/
│   │   ├── llm.py                   # Groq LLM integration
│   │   ├── tts.py                   # Text-to-speech
│   │   └── voice_pipeline.py        # Voice coaching pipeline
│   ├── config/workout_config.py     # Exercise options & prompts
│   ├── persistence/exercise_repository.py  # SQLite database
│   ├── state/session_defaults.py    # Session state management
│   ├── tracking/metrics.py          # Rep/set tracking logic
│   ├── ui/style_loader.py           # CSS and font injection
│   └── vision/exercise_video_processor.py  # Video processing
└── static/
    ├── style.css                    # Custom styles
    └── AdobeClean.otf               # Custom font

Setup

# Create conda environment
conda create -n gymcoach python=3.11 -y
conda activate gymcoach

# Install dependencies
pip install -r requirements.txt

# Add your Groq API key
# Create a .env file with: GROQ_API_KEY=your_key_here

# Run the app
streamlit run main.py

Environment Variables

Variable	Description
GROQ_API_KEY	API key from console.groq.com

Built By

Tejas Manoj — Data & ML Engineer

• LinkedIn
• GitHub