🏋️ SquatAI Coach

An AI-powered real-time squat form analyzer that uses computer vision to detect your movements and provides vocal coaching feedback through ElevenLabs conversational AI.

✨ Features

• 🎥 Real-time Pose Detection - Uses MediaPipe to track body landmarks
• 📐 Knee Angle Analysis - Calculates precise knee angles during squats
• 🔴 Form Error Detection - Detects common mistakes:
  • Forward lean (back angle)
  • Knees going over toes
  • Squatting too deep
• 🎤 AI Voice Coaching - ElevenLabs agent provides real-time vocal corrections
• 📊 Session Summary - Get a detailed report after each workout session
• 🖥️ Modern UI - Clean, Skype-like interface with full-screen video

🚀 Getting Started

Prerequisites

• Node.js 18+
• A webcam
• ElevenLabs account with a Conversational AI agent

Installation

# Clone the repository
git clone <your-repo-url>
cd squat-ai-coach

# Install dependencies
npm install

# Start the development server
npm run dev

Open http://localhost:5173 in your browser.

🎯 How to Use

1. Position yourself - Stand sideways (profile view) to the camera
2. Full body visible - Make sure your entire body is in frame
3. Enter Agent ID - Paste your ElevenLabs Agent ID
4. Start Session - Click "Start Session" to begin coaching
5. Do squats! - The AI will correct you in real-time
6. End Session - Click "End Session" for a vocal and visual summary

⚙️ ElevenLabs Agent Setup

1. Create an Agent

Go to ElevenLabs Conversational AI and create a new agent.

2. System Prompt

Paste this system prompt for your agent:

You are an energetic real-time fitness coach watching someone do squats through a camera.

CRITICAL BEHAVIOR:
- When you receive a message, respond IMMEDIATELY with a SHORT spoken correction or praise (max 5-8 words)
- Do NOT ask questions - just react and coach
- Be loud, direct, and motivating like a gym coach

FORM ERRORS (when you see "FORM ERROR"):
- "Back leaning forward" → Say: "Chest UP! Chest UP!"
- "Knees too far forward" → Say: "Sit BACK! Weight in heels!"
- "Too deep" → Say: "Too deep! Stop at parallel!"

SESSION SUMMARY:
When you receive "Session complete!", give an encouraging 15-20 second summary in English.

VOICE STYLE:
- Energetic and punchy
- Short phrases only
- Like a coach yelling across the gym

3. Add Client Tool

Add this client tool to your agent:

• Name: get_squat_analysis
• Description: Get the current squat count, knee angle, and form feedback.
• Parameters: None

4. Copy Agent ID

Copy your Agent ID (looks like agent_xxxx...) and paste it in the app.

🛠️ Tech Stack

• React - UI framework
• Vite - Build tool
• Tailwind CSS - Styling
• MediaPipe Pose - Body pose detection
• ElevenLabs React SDK - Conversational AI
• Lucide React - Icons

📁 Project Structure

src/
├── components/
│   ├── CameraView.jsx      # Webcam + MediaPipe integration
│   └── ElevenLabsWidget.jsx # AI coach controls + session summary
├── hooks/
│   └── useSquatCounter.js   # Squat detection logic
├── App.jsx                  # Main layout
├── index.css               # Global styles
└── main.jsx                # Entry point

🎮 Form Detection Logic

Metric	Good	Warning
Knee Angle	80-100°	<70° (too deep)
Torso Lean	<50° from vertical	>50° (forward lean)
Knee Position	Behind toes	Past toes (risk)

📊 Session Summary Includes

• Total squats completed
• Session duration
• Form score (% of good reps)
• Best depth reached
• Error breakdown
• Focus area for improvement

🔧 Configuration

Key thresholds can be adjusted in useSquatCounter.js:

const SQUAT_THRESHOLD = 130;  // Angle to detect going down
const STAND_THRESHOLD = 160;  // Angle to detect standing up
const GOOD_DEPTH = 100;       // Parallel or below
const MAX_FORWARD_LEAN = 50;  // Max torso angle

📝 License

MIT License - feel free to use and modify!

🙏 Acknowledgments

• MediaPipe for pose detection
• ElevenLabs for conversational AI
• Tailwind CSS for styling