🩸 Complete NightscoutAI transformation: Fix critical bugs, add modern UI/UX, mmol/L support, continuous training, anomaly detection, and robust chart rendering

README.md

@@ -0,0 +1,114 @@
+# 🩸 NightscoutAI - Blood Glucose Prediction Dashboard
+
+An AI-powered blood glucose prediction tool that trains entirely in your browser using Nightscout CGM data.
+
+![NightscoutAI Dashboard](https://via.placeholder.com/800x400?text=NightscoutAI+Dashboard)
+
+## ✨ Features
+
+- **🤖 Real-time AI Training**: LSTM neural network trains on your Nightscout data directly in the browser
+- **📊 Interactive Charts**: Beautiful visualizations of actual vs predicted blood glucose levels
+- **💾 Model Persistence**: Save and load trained models locally in your browser
+- **🎮 Demo Mode**: Test the application with sample data when Nightscout API is unavailable
+- **📱 Responsive Design**: Works on desktop, tablet, and mobile devices
+- **🔐 Privacy-First**: All data processing happens locally - no data sent to external servers
+- **⚡ Real-time Predictions**: Get instant blood glucose predictions based on historical patterns
+
+## 🚀 Quick Start
+
+1. **Open the Application**: Simply open `index.html` in a modern web browser
+2. **Enter Your Nightscout URL**: Input your Nightscout site URL (e.g., `https://yoursite.herokuapp.com`)
+3. **Fetch & Train**: Click the "🔄 Fetch & Train" button to download data and train the AI model
+4. **View Predictions**: See your current BG and AI-predicted next reading
+
+### Demo Mode
+
+If you don't have access to a Nightscout site or want to test the application:
+1. Click the "🎮 Demo Mode" button
+2. The app will use sample blood glucose data to demonstrate functionality
+
+## 🔧 Technical Details
+
+### AI Model Architecture
+- **Model Type**: LSTM (Long Short-Term Memory) Neural Network
+- **Input Features**: Blood glucose, insulin doses, carbohydrate intake
+- **Sequence Length**: 10 time points for pattern recognition
+- **Training**: Online learning with each data update
+
+### Technologies Used
+- **TensorFlow.js**: Machine learning in the browser
+- **Chart.js**: Interactive data visualizations
+- **Vanilla JavaScript**: Lightweight, no framework dependencies
+- **HTML5/CSS3**: Modern responsive design
+
+### Data Processing
+- Fetches data from Nightscout API (`/api/v1/entries.json`)
+- Normalizes blood glucose values (40-400 mg/dL range)
+- Filters invalid readings and sorts by timestamp
+- Creates sliding windows for sequence-based learning
+
+## 📊 Statistics Dashboard
+
+The application provides comprehensive training statistics:
+- **Average Loss**: Model training loss (lower is better)
+- **Prediction Error**: Average prediction accuracy in mg/dL
+- **Training Steps**: Total number of training iterations
+- **Data Points**: Total CGM readings processed
+
+## 💡 Usage Tips
+
+1. **Data Quality**: More historical data generally leads to better predictions
+2. **Regular Updates**: Train periodically with new data for improved accuracy
+3. **Model Saving**: Save your trained model to avoid retraining each session
+4. **Browser Compatibility**: Works best in Chrome, Firefox, Safari, and Edge
+
+## ⚠️ Important Notes
+
+- **Not Medical Advice**: This tool is for educational/research purposes only
+- **Supplement, Don't Replace**: Should complement, not replace, medical monitoring
+- **Data Privacy**: All processing happens locally in your browser
+- **Internet Required**: Needs internet connection to fetch Nightscout data (except demo mode)
+
+## 🔧 Configuration
+
+### Adjustable Parameters (in code)
+```javascript
+const SEQ_LEN = 10;        // Sequence length for predictions
+const LSTM_UNITS = 16;     // Neural network complexity
+const BG_MIN = 40;         // Minimum BG value for normalization
+const BG_MAX = 400;        // Maximum BG value for normalization
+```
+
+## 🐛 Troubleshooting
+
+**Issue**: External libraries not loading
+- **Solution**: The app includes fallback CDNs and demo mode for offline testing
+
+**Issue**: No data from Nightscout
+- **Solution**: Verify your Nightscout URL is correct and accessible
+
+**Issue**: Poor prediction accuracy
+- **Solution**: Ensure sufficient historical data (at least 100+ readings)
+
+## 🤝 Contributing
+
+Contributions are welcome! Some areas for improvement:
+- Enhanced model architectures
+- Additional input features (exercise, stress, sleep)
+- Mobile app version
+- Advanced visualization options
+- Multi-step ahead predictions
+
+## 📄 License
+
+This project is open source. Please ensure compliance with medical device regulations in your jurisdiction.
+
+## 🙏 Acknowledgments
+
+- Nightscout community for the open CGM platform
+- TensorFlow.js team for browser-based machine learning
+- Chart.js for excellent visualization capabilities
+
+---
+
+**⚠️ Medical Disclaimer**: This software is not intended for medical diagnosis or treatment. Always consult healthcare professionals for medical decisions.