Contact Information

160 Robinson Rd, #14-04
Sit Rd, SBF Center, Singapore 068914

Defog, a Singapore-based analytics firm powered by AI, has recently secured $2.2M in seed funding. This funding round, led by Script Capital and Y Combinator, saw participation from Hike Ventures, Pioneer Fund, and several angel investors.

  • Funding Utility: Fast-track the development of SQLCoder, their open-source large language model (LLM). This model is designed for deployment on corporate databases to enable swift data analysis. It allows employees to raise queries in plain English using structured data, with SQLCoder generating answers up to 80% faster than traditional methods.
  • More Developments: Speed up development of their Agents platform. This platform will enable enterprises to ask even more complex data-related questions using AI. The Agents platform tests various data models, parameter settings, and data slices to find the best solutions, thus automating many trial-and-error tasks required for statistical analysis.
  • Unique Selling Point: Despite the US being a competitive market, with major players like Google and Microsoft’s OpenAI, Defog stands its ground due to its unique selling point: the ability to tailor its LLM for each client. This gives companies a significant advantage, offering them complete access to the LLM’s operations, allowing them to leverage it according to their needs.
  • Monetization Process: Fine-tuning the SQLCoder model for each client, considering their unique database structures and definitions. Assisting in the deployment of SQLCoder as either a cloud-hosted or on-premise solution, and pricing based on the volume of queries generated from the model.

According to Defog’s CEO, Medha Basu, task-specific models like SQLCoder are more cost-effective than generalist models. Smaller task-specific models, such as Defog’s 7 billion-parameter SQLCoder, are more affordable to host than larger, more powerful models like GPT4, which has more than a trillion parameters.

client banner image

Share:

Leave a Reply

Your email address will not be published. Required fields are marked *