Gretel Navigator

Gretel Navigator is a compound AI system that can create, edit and augment tabular data, letting people build datasets step by step from scratch. The tool is designed to improve data quality and security by producing data that's realistic enough to be used for AI training, testing, demonstrations and evaluation.

Gretel Navigator comes in two modes: Create and Edit. The Create mode produces realistic data when none exists, while the Edit mode lets people modify, augment and fill in gaps in existing data with SQL or natural language prompts. That can help overcome difficulties in creating data in the first place and managing it afterward.

Some examples of Gretel Navigator use cases:

• Training foundation models: Creating high-quality training data for Large Language Models (LLMs).
• Fine-tuning LLMs: Training LLMs for new tasks or domains for generative AI use.
• Creating evaluation datasets: Creating synthetic question-truth pairs to evaluate RAG models.
• Testing and evaluating ML models: Protecting sensitive data when evaluating publicly available ML models.
• Data augmentation for ML training: Improving real-world data to train more capable AI applications.
• Personalizing product demos: Creating product demos that are tailored to prospect business needs.

Gretel Navigator has been used to create open-source datasets, including the world's largest Text-to-SQL dataset and synthetic financial documents with PII labels, both released under Apache 2.0.

For large-scale use, Gretel Navigator also offers a real-time inference API that can be built directly into data services with an SDK. That lets people create custom datasets by adding new correlated data, filling in gaps or adding regional attributes.

Companies like Ernst & Young and Databricks have benefited from Gretel Navigator, including better data quality, lower costs and faster product development.

Published on June 14, 2024