Exclusive: Y Combinator leads $2.2m round in SG-based AI analytics firm

Defog, an AI-powered analytics firm based in Singapore, has raised US$2.2 million in a seed round led by Script Capital and Y Combinator. Hike Ventures and Pioneer Fund as well as several angel investors also joined the round.

With its latest fundraise, Defog plans to accelerate the development of SQLCoder, its open-source large language model (LLM).

The startup’s model is designed to be deployed on a company’s database and facilitate rapid data analysis. Employees can use structured data to pose questions in simple English, and SQLCoder can generate answers 80% faster than manual methods.

Defog is also looking to expedite the development of its Agents platform, “which will empower enterprises to ask even more complex questions using AI,” co-founder and CEO Medha Basu told Tech in Asia.

Defog co-founders Medha Basu (left) and Rishabh Srivastava | Photo credit: Defog

By testing different models, parameter settings, and data slices to find the best solutions, Defog’s Agents can automate many of the trial-and-error tasks required for statistical analysis.

According to Basu, Defog is now focusing on the US since that market is adopting AI tech faster than Southeast Asia. “We didn’t target the US specifically. It happened organically because of their advancement in AI. I think being part of Y Combinator also helped us meet some of these customers,” she says.

But with major players like Google and Microsoft’s OpenAI, the US is a tough market. What sets Defog apart, Basu notes, is its ability to customize its LLM to suit a client’s needs, giving them “an incredibly important and powerful competitive edge.”

Companies want complete access to an LLM’s inner workings and have the ability to leverage them as they see fit, she adds. In comparison, having a model that gives the impression of putting information into a black box via an API doesn’t give companies control over the process, says Basu.

Path to monetization

“The first step is fine-tuning the model for them,” says Basu. Defog tweaks its SQLCoder model for each customer, factoring in the nuances of their database structures and specific definitions.

Basu points out how even seemingly straightforward tasks, such as understanding customer churn or calculating the cost of an item, can vary for each customer.

Defog also helps with deploying SQLCoder as either a cloud-hosted or on-premise solution – the second step in the company’s monetization.

“The final layer, of course, is based on the query volume,” says Basu. This varies based on the volume of queries that customers can generate from the model.