Battery Smart Voice Assistant - NLU Model

Intent classification and entity extraction for Battery Smart's multilingual voice assistant (Hindi, English, Hinglish).

🎯 Features

• 23 Intent Classes: Battery status, swap stations, payments, subscriptions, etc.
• Multilingual Support: Hindi, English, and Hinglish
• MuRIL Model: Google's Multilingual Representations for Indian Languages
• Augmented Dataset: 1,230+ training examples
• High Performance: 70%+ accuracy on test set

📊 Model Performance

Intent Classifier

• Model: google/muril-base-cased
• Training Examples: 1,230 (augmented from 342)
• Test Accuracy: 70%+ (improved from 48.6%)
• Supported Intents: 23 classes

Dataset Statistics

Training: 1,230 examples
Validation: 153 examples
Test: 155 examples
Languages: Hindi, English, Hinglish

🚀 Quick Start

Installation

# Clone repository
git clone <your-repo-url>
cd model_proj

# Install dependencies
pip install -r requirements.txt

# Configure Python environment (if using venv)
python -m venv nlu_env
source nlu_env/bin/activate  # On Windows: nlu_env\Scripts\activate
pip install -r requirements.txt

Training

# Run data augmentation
cd src
python3 improve_and_retrain.py

# Train intent classifier
python3 train_intent_classifier.py

Testing

# Interactive testing
cd src
python3 test_intent_model.py

📁 Project Structure

model_proj/
├── config/              # Intent, entity, domain configurations
│   ├── intents.yaml
│   ├── entities.yaml
│   └── domain.yaml
├── data/
│   ├── processed/       # Training data
│   │   ├── train_augmented.json
│   │   ├── val_augmented.json
│   │   └── test_augmented.json
│   └── raw/            # Original data
├── models/             # Trained models
│   └── intent_classifier/
├── src/
│   ├── train_intent_classifier.py  # Training script
│   ├── test_intent_model.py        # Testing script
│   ├── improve_and_retrain.py      # Data augmentation
│   ├── training/                    # Training utilities
│   ├── nlu/                        # NLU pipeline
│   └── api/                        # FastAPI endpoints
└── requirements.txt

🎓 Supported Intents

1. Battery & Charging

   • check_battery_status - Current charge level
   • check_battery_range - Remaining range

2. Swap Stations

   • find_swap_station - Locate nearest station
   • check_station_availability - Station status
   • station_directions - Get directions

3. Subscriptions & Payments

   • check_subscription - Plan details
   • subscription_renewal - Renew plan
   • upgrade_subscription - Upgrade plan
   • make_payment - Process payment
   • payment_inquiry - Payment status
   • payment_history - Transaction history

4. Usage & History

   • swap_history - Past swaps
   • swap_count - Total swaps
   • check_ride_stats - Ride statistics

5. Support

   • report_issue - Report problems
   • talk_to_agent - Human support
   • help_general - General help

6. Conversation

   • greet, bye, thank, affirm, deny

🔧 Configuration

Training Hyperparameters

num_epochs: 15
batch_size: 16
learning_rate: 3e-5
early_stopping_patience: 5

Data Augmentation

• Hinglish variations
• Polite form additions
• Class balancing (30+ samples per class)
• Text transformations

📈 Improvement Roadmap

• Data augmentation
• Balanced dataset
• Improved hyperparameters
• Entity extraction training
• Complete NLU pipeline
• API deployment
• Multi-turn dialogue support

🤝 Contributing

See TRAINING_GUIDE.md for details on training on AWS Bedrock and contributing improvements.

📄 License

[Your License Here]

📞 Contact

[Your Contact Information]