Wellby

Work well. Rest well. Be well.

Wellby is a friendly AI-powered wellbeing companion for tech industry workers. It tracks how you work, learns your baseline pace, estimates burnout risk, and nudges you toward restorative breaks with a breathing exercise and a built-in game lounge.

Stack

• Frontend: React + Tailwind CSS
• Backend: Node.js + Express
• Local AI services:
  • Local burnout proxy model trained on the Syncora synthetic productivity dataset
  • Background EAR fatigue detection inspired by Aeidle/EAR-Fatigue-Detection

Included features

• Personalized onboarding with name, work setup, seniority, work hours, and favorite break game
• Friendly dashboard with live task tracking and actions-per-minute monitoring
• Adaptive flow engine that builds a baseline over the first 3 sessions
• Burnout risk meter with warm status copy and weekly trend chart
• Three-tier burnout alert system with toast, banner, modal-style escalation, and snoozes
• Opt-in fatigue detection polling with a local-only privacy notice
• Break Mode with a 60-second mindful breathing moment
• Built-in Snake, Chess, Tic Tac Toe, and simplified UNO mini-games
• Local persistence for onboarding, burnout history, sessions, and game scores

Local development

1. Clone this repo.
2. Clone the burnout prediction service:
   • Included training dataset: syncora-ai/Synthetic-AI-Developer-Productivity-Dataset
   • In this repo install Python dependencies: pip install -r requirements.txt
   • Train the local model: python train_burnout_model.py
   • Run the bundled burnout service: python flask_api.py --port 5001
3. Set up the fatigue detection service:
   • Reference algorithm: Aeidle/EAR-Fatigue-Detection
   • In this repo install the local service dependencies: pip install -r requirements.txt
   • Run the bundled background service: python fatigue_service.py --port 5002
   • The service keeps the webcam monitor running locally in the background and exposes /start, /stop, and /status on port 5002
4. In this repo:
   • npm install
   • npm run dev
5. Open http://localhost:3000
6. Complete onboarding and start your first Wellby session.

Scripts

• npm run dev starts the Express server on port 3000 and mounts the Vite React frontend in middleware mode.
• npm run build builds the React client into dist/.
• npm start serves the production build from Express.

API surface

• GET /api/health
• POST /api/burnout/predict
• GET /api/fatigue/status
• POST /api/fatigue/start
• POST /api/fatigue/stop

The backend proxies the two Python microservices and falls back gracefully if either service is unavailable during local setup.

Burnout model notes

• The bundled burnout service trains a local RandomForestRegressor against the Syncora synthetic developer productivity dataset.
• The service keeps the existing /predict contract and maps Wellby's payload into the trained feature space used by the model.
• Trained model artifacts are stored in artifacts/.

Ethics and data use

• The Syncora dataset is synthetic and is described by its repo as carrying zero risk of exposing real employee data.
• The dataset repo frames it for research, education, and experimentation rather than employee surveillance.
• Wellby treats the trained score as a wellbeing estimate only; it should not be used for HR discipline, ranking, or diagnosis.

NPM packages used

• chess.js
• chessboard.js package included in package.json
• chart.js
• react-hot-toast
• framer-motion

Notes

• Weekly burnout history, session summaries, break logs, and game scores are stored in localStorage.
• The fatigue detection toggle starts the local EAR detector in the background when the user opts in and stops it when disabled.
• After each completed break, Wellby applies a -0.1 reducing factor to the next burnout score, with a floor of 0.0.