Jivani Models

A collection of natural language processing models used for the Jivani AI assistant.

Command Recognition Model

The command recognition system uses a fine-tuned T5 model to parse natural language commands into structured actions.

Dataset

The dataset used for training contains examples of:

• Task management (create, update, delete tasks)
• Goal tracking (add, complete goals)
• Habit tracking (add habits)
• Note taking (add, update, delete notes)
• Reminders (set, delete reminders)
• Financial tracking (log expenses, income)
• List management (wishlist, shopping list)

Dataset files:

• dataset/train_dataset.json - Training data
• dataset/validation_dataset.json - Validation data
• dataset/test_dataset.json - Test data

Model Details

The model is based on T5 (Text-to-Text Transfer Transformer) and trained to:

1. Parse natural language commands
2. Extract relevant entities and parameters
3. Generate structured JSON outputs with action types and attributes

Model files are stored in command-recognition/weights/

Training

The model training code is available in command-recognition/model-training/t5Final.ipynb