Defog.ai

Defog.ai, previously Data Narratives, operates from Mountain View, CA, and employs a team of five. The company is part of the B2B analytics sector and participated in the Y-Combinator W23 batch. Defog.ai serves clients across the United States and Canada, offering both remote and partly remote services. The company’s core technology, SQLCoder, is an open-source model that enables business users to query data using everyday language and is capable of being fine-tuned and deployed on-premises. SQLCoder powers solutions that accelerate data analyses and automate complex workflows in SQL, Python, and R, integrating privacy-centric features with any SQL database or data warehouse. Defog.ai’s technology can decompose intricate, multi-part questions into executable tasks, automatically visualizing data and generating editable reports. The AI models are customized to align with user-specific database schemas and business logic, offering both cloud-based services and on-premises deployment options. Notably, SQLCoder-34B, a version of their model, surpasses OpenAI’s gpt-4 in text to SQL generation and demonstrates over 99% accuracy in text to SQL conversions when fine-tuned on individual database schemas. Additionally, Defog.ai provides SQL-Eval, a framework for evaluating the correctness of LLM-generated SQL, ensuring compatibility with all major SQL databases and data warehouses.