Gretel Navigator Tabular generates production-quality synthetic data optimized for AI and machine learning development from prompts, schema definitions, or seed examples. Unlike single-LLM approaches to data generation, Navigator Tabular employs a compound AI architecture specifically engineered for synthetic data, combining top open-source SLM models fine-tuned across 10+ industry domains. This purpose-built system creates diverse, domain-specific datasets at scales of hundreds to millions of examples while preserving complex statistical relationships and offering increased speed and accuracy compared to manual data creation.

Key Features

• Natural language interface to specify data requirements
• Schema-based data generation
• Real-time and streaming data generation
• Dataset augmentation and modification
• Structured data supported as LLM inputs and outputs

Top Use Cases

• Creating synthetic data for LLM training and fine-tuning
• Generating evaluation datasets for AI models and RAG systems
• Augmenting limited training data with diverse synthetic samples
• Creating realistic PII/PHI data for model testing

Documentation and Resources

• Example Notebooks

Input Examples

Natural Language Prompts

Generate customer bank transaction data with the following columns:
- customer_name: Full names in Western format
- transaction_date: Dates within the last 30 days
- transaction_amount: Dollar amounts between $1-$10,000
- transaction_type: Either 'debit' or 'credit'
- transaction_category: Common banking categories like 'dining', 'retail', 'utilities'
- account_balance: Running balance after each transaction

Schema-Based Input

CREATE TABLE transactions (
    customer_name VARCHAR(100),
    customer_id CHAR(8),
    transaction_date DATE,
    transaction_amount DECIMAL(10,2),
    transaction_type VARCHAR(6),
    transaction_category VARCHAR(50),
    account_balance DECIMAL(10,2)
);