An AI-powered personal training app that provides real-time form coaching, injury risk scoring, rep counting, and recovery tracking — all running in the browser.

• Real-time pose detection — MediaPipe Pose tracks 33 body landmarks via webcam at 30fps
• Exercise rep counting — angle-based detection for squat, pushup, lunge, deadlift, shoulder press, curl-up, and bicep curl
• AI form coaching — GPT-4o analyzes live workout frames and gives specific, actionable cues
• Injury risk scoring — blended local geometry score + OpenAI vision score, updated every 15 seconds
• Warmup quality gate — rates warmup form 0–100 before allowing the main workout
• AI cooldown generation — GPT-4o-mini generates a personalized cooldown plan based on the session
• Set counter — press S or Space to mark a new set; history shown in sidebar
• Session summary — rep counts, avg/peak risk score, form suggestions, feel rating
• Recovery logging — daily sleep, energy, mood, soreness-per-muscle, RPE
• Friend squads — search users by display name, send/accept friend requests, shared streaks
• Prime Intelligence — group AI briefing with leaderboard, duo links, and warden's directive
• OpenAI TTS — spoken coaching cues (alloy voice), falls back to Web Speech API
• Streak tracking — consecutive workout day counter synced to Firestore

src/
├── components/
│   ├── RootRedirect.tsx       # Auth guard — redirects to /auth or /onboarding
│   └── BodySorenessMap.tsx    # Interactive per-muscle soreness selector
├── hooks/
│   ├── usePoseDetection.ts    # MediaPipe webcam loop, frame buffer, skeleton overlay
│   └── useRepCounter.ts       # Angle/position based rep counting for all exercises
├── lib/
│   ├── firebase.ts            # Firebase app initialization
│   ├── firebaseHelpers.ts     # Firestore CRUD — sessions, logs, users, friends
│   ├── firestoreUser.ts       # Auth state helpers, uid resolution
│   ├── formAnalysis.ts        # OpenAI vision analysis, TTS, cooldown generation
│   ├── primeIntelligence.ts   # Group AI coach — briefings, leaderboards, duo links
│   ├── localUserId.ts         # Offline/anonymous user id fallback
│   ├── recoveryLogLocal.ts    # localStorage wrapper for recovery logs
│   └── recoveryMuscles.ts     # Muscle group definitions
├── pages/
│   ├── AuthPage.tsx           # Email/password sign-in and sign-up
│   ├── OnboardingPage.tsx     # First-time profile setup
│   ├── HomePage.tsx           # Dashboard — streak, recent sessions, quick actions
│   ├── WorkoutPage.tsx        # Live workout — camera, reps, risk score, AI feedback
│   ├── SessionSummaryPage.tsx # Post-workout summary with stats and suggestions
│   ├── CooldownPage.tsx       # Timed cooldown exercises (AI-generated)
│   ├── RecoveryLogPage.tsx    # Daily health and soreness logging
│   ├── ProfilePage.tsx        # Profile editor, OpenAI key management, sign out
│   ├── FriendsPage.tsx        # Squad search, friend requests, Prime Intelligence
│   └── PipelineTestPage.tsx   # Internal test page for AI pipeline
├── stores/
│   └── workoutStore.ts        # Zustand store for active session state
├── types/
│   └── index.ts               # All shared TypeScript interfaces
└── App.tsx                    # Route definitions + auth sync on boot

• Node.js 18+
• A Firebase project with Authentication (email/password) and Firestore enabled
• An OpenAI API key (optional — basic mode works without it)

git clone https://github.com/your-org/IntoYourPrime.git
cd IntoYourPrime
npm install

Create a .env file at the project root:

VITE_FIREBASE_API_KEY=...
VITE_FIREBASE_AUTH_DOMAIN=your-project.firebaseapp.com
VITE_FIREBASE_PROJECT_ID=your-project-id
VITE_FIREBASE_STORAGE_BUCKET=your-project.firebasestorage.app
VITE_FIREBASE_MESSAGING_SENDER_ID=...
VITE_FIREBASE_APP_ID=...

# Optional — users can also enter their key in the Profile page
VITE_OPENAI_API_KEY=sk-proj-...

npx firebase-tools login
npx firebase-tools deploy --only firestore:rules --project your-project-id

Or paste the contents of firestore.rules into Firebase Console → Firestore → Rules and publish.

npm run dev

The firestore.rules file at the project root defines the access policy:

• users/{uid} — any authenticated user can read (enables friend search); only the owner can write
• friendConnections/{connId} — only participants can read/update; sender can create
• sessions, dailyLogs — owner only
• activityFeed — any authenticated user can read; owner can create

Rep counting is fully local (no API calls). Each exercise uses a different anatomical signal:

All signals use EMA smoothing (α=0.2), a 1-second calibration window, and a 1.2-second debounce between reps.

When an OpenAI API key is present:

• Form analysis runs every 15 seconds (first call at 5s), sending 3–5 webcam frames to GPT-4o
• Suggestions are specific second-person cues: "Your left knee is caving inward — press it out over your pinky toe."
• Risk score is blended: 60% AI vision score + 40% local geometry score
• Cooldown is generated by GPT-4o-mini after the main session ends
• TTS reads the top suggestion aloud (OpenAI tts-1, alloy voice)

Without an API key, local geometry scoring still provides risk feedback and canned coaching suggestions rotate every 10 seconds.

npm run dev        # Start dev server (Vite HMR)
npm run build      # Type-check + production build
npm run lint       # ESLint
npm run preview    # Preview production build locally