Roadmap

Current state of Posey and planned improvements.

Current state (v0.1)

  • Google Sign-In + Firebase Auth
  • User onboarding (AI-powered adaptive questionnaire)
  • Training plan generation via Gemini + Firestore
  • Training plan UI (detail view, week view, regenerate)
  • Camera capture with front/back toggle
  • Two pose detection modes: QuickPose (default) and Apple Vision
  • Real-time 2D pose detection (19 keypoints, skeleton overlay)
  • Joint angle calculation (9 angles)
  • Gemini Live API integration with bidirectional streaming
  • Voice coaching output and input (PCM audio)
  • User can interrupt agent mid-response (barge-in)
  • Text subtitle overlay for AI responses
  • Auto exercise detection via AI
  • Workout session lifecycle (start/stop/reconnect)
  • Rep counting (hysteresis-based, QuickPose threshold counters)

Short term

Workout history & logging

  • Save workout sessions with timestamps, duration, detected exercises
  • Store AI coaching messages per session
  • Local persistence with SwiftData
  • Post-workout summary screen

Voice input improvements

  • Ask questions mid-workout — basic support done
  • Push-to-talk mode (optional)
  • Echo cancellation when using speaker

Improved pose overlay

  • Color-code joints by confidence level
  • Highlight joints involved in current exercise
  • Show angle values on screen for key joints
  • Draw angle arcs at joint vertices

Settings screen

  • AI voice selection
  • Coaching style preference
  • Camera and update frequency preferences
  • Language selection (70+ languages)

Medium term

3D pose detection

  • 17 keypoints with depth for better bench press, overhead press analysis

Workout plans & guided sessions

  • Pre-built templates (Push/Pull/Legs, Full Body, HIIT)
  • AI guides through plan, tracks rest periods
  • Rest timer with haptic feedback

Form score system

  • Quantitative scoring (0-100) per rep
  • Visual feedback: green/yellow/red glow based on form quality

Exercise library

  • Database of exercises with ideal angle ranges
  • "Ideal vs actual" overlay

Offline mode

  • Cache AI coaching rules locally
  • On-device pose analysis when offline

Long term

  • Social features — share summaries, challenges, leaderboards
  • Video recording & review with pose overlay baked in
  • Multi-person detection for group workouts
  • Apple Watch integration — heart rate, haptic cues, HealthKit
  • Advanced AI — injury risk, periodization, nutrition tips
  • On-device AI (Apple Intelligence / FoundationModels)

Technical debt & improvements

Performance

  • Profile CPU/GPU usage
  • Adaptive frame skipping for older devices
  • Optimize JPEG compression

Reliability

  • Network connectivity monitoring
  • Exponential backoff for Gemini reconnection
  • Graceful degradation when AI disconnects
  • Audio session interruption handling

Code quality

  • Unit tests for PoseAnalyzer
  • UI tests for workout flow
  • Integration tests for Gemini service

Security

  • Firebase App Check
  • Remote Config for model name
  • Rate limiting on Gemini sends