We are excited to announce Defog’s $2.2M funding round for building open-source LLMs for data analysis. The round was led by Script Capital and Y Combinator, with participation from Hike Ventures, Pioneer Fund, and notable angel investors including Dharmesh Shah (Co-Founder of HubSpot), and Divya Bhat (former Y Combinator Visiting Partner).
Defog helps enterprises use AI for faster data analysis. Our open-core LLMs can be deployed and fine-tuned on-prem, and outperform generalist models like GPT-4 for data analysis tasks, like text to SQL conversion.
The core of Defog is powered by our open-source large language models, SQLCoder - the world’s best models for converting natural language questions into SQL queries. In just three months, SQLCoder models have been downloaded over 50,000 times on Hugging Face and have been starred 1800 times on Github.
With Defog, employees can ask questions that require complex analyses in plain English, and have them be answered in minutes instead of hours or days.
Building open-source LLMs to unlock enterprise productivity
Our latest 34B model beats gpt-4-turbo and gpt-4 on the sql-eval benchmark for out-of-training-set schemas. When fine-tuned to an enterprise’s database schema, we get 99+% accuracy.
Data scientists and business leaders at large companies waste 1000s of hours every month trying to get data to make business decisions. Today, our enterprise offering is being used by healthcare, finance and analytics companies who are leveraging AI for faster data analyses. A publicly-listed US company has deployed Defog to a 100-person department and cut analysis time by 80%, with an estimated savings of 2,500 hours per month for its data science team.
A personal AI analyst for every employee
Our vision is to help enterprises provide an AI data analyst for every employee, while giving organisations complete control over their AI.
We are particularly excited about what our AI Agents will enable next. Our latest offering Defog Agents lets users to ask more complex "why" questions. Agents understands users’ intent, explores hypotheses, writes code to extracts the right data and run analyses, and returns results as tables, charts and reports. As a result, complex analyses that took months are being done in 2-3 days.
← More blogs