Behave

AI-powered behavioral coaching for remote professionals. Uses on-device computer vision and speech analysis to help you improve posture, facial expressions, habits, and communication in real-time.

Architecture

• iOS app (Swift/SwiftUI) — all detection runs on-device via Apple Vision framework (Neural Engine)
• Backend API (FastAPI) — auth, data sync, LLM coaching via Claude
• Deployment — Shelob on kieleth-sandbox (behave.kieleth.com)

Detection pipeline

Detector	Framework	Output
Body pose	VNDetectHumanBodyPoseRequest	19 joint points
Face landmarks	VNDetectFaceLandmarksRequest	76 landmarks
Hand pose	VNDetectHumanHandPoseRequest	21 landmarks/hand
Speech	SFSpeechRecognizer (on-device)	Real-time transcription

Project structure

behave/
├── legacy/          # Original Python code (2014)
├── ios/             # SwiftUI iOS app
│   ├── Behave/
│   │   ├── Detection/       # Vision framework detectors
│   │   ├── Classification/  # Behavior classifiers
│   │   ├── Enforcement/     # Alert/rule engine
│   │   ├── Views/           # SwiftUI views
│   │   └── Data/            # Models, persistence
│   └── project.yml          # xcodegen spec
├── backend/         # FastAPI backend
│   ├── app/
│   │   ├── auth/            # Apple Sign-In + JWT
│   │   ├── models/          # SQLAlchemy models
│   │   ├── routers/         # API endpoints
│   │   └── services/        # Claude proxy
│   └── Dockerfile
├── docker-compose.yml
└── shelob.yml

Development

iOS app

cd ios
xcodegen generate
open Behave.xcodeproj

Backend (local)

cd backend
pip install -r requirements.txt
uvicorn app.main:app --reload --port 8300

Deploy

shelob deploy_project behave