Salesforce Ventures Joins Korean AI Startup Datumo’s Series B
Datumo’s founders, clockwise from top left: Yuntaek Lee, Haseung Jeon, Michael Hwang, David Kim and Munhwi Jeon.
Courtesy of Datumo
Seoul-based Datumo is the latest AI startup to raise cash from investors, closing a $15.5 million Series B round to keep up with growing demand for its model training datasets and evaluation, the company said.
Investors in the latest round include Salesforce Ventures, Palo Alto-based ACVC Partners, Japan’s SBI Investment, KB Investment, Shinhan Venture Investment, Kiwoom Investment and Moorim Capital, the finance arm of Korean paper manufacturer Moorim. Datumo declined to disclose the valuation of the round. The startup’s new funding brings its total to about $25 million.
The investment is Salesforce Ventures’ first in an AI startup based in South Korea. “In an era when AI safety concerns are the largest hurdle to widespread adoption, Datumo provides a generative AI evaluation solution that enables organizations to deploy AI-powered services across any domain at scale confidently,” Salesforce Ventures said in a statement.
Besides financial support, Salesforce’s backing can provide Datumo with several benefits, including the possibility of working with the company on its AI agents and the software giant’s customers. “Generative AI is advancing at breakneck speed, with new services launching daily on top of foundational models,” Salesforce Ventures added in the statement. “But for every company building with LLMs, one critical challenge remains: ensuring the reliability and safety of AI-generated outputs.”
Founded in 2018, Datumo, previously known as SelectStar, provides tools that can check the reliability of LLM responses and evaluate their performance to mitigate mistakes before deployment. “It’s just how foundation models work. They are probabilistic by nature, meaning they don’t give deterministic results, so their output can change based on very small variation in input or prompt,” explains David Kim, cofounder and CEO of Datumo, in a video interview. “So it makes quality assurance extremely hard.”
Kim, who was on Forbes’ 30 Under 30 Asia list in 2021, says he’s seeing growing demand from companies in regulated industries like finance. “AI in regulated industries cannot just be good enough,” he says. “It needs to be safe and align with the domain-specific expectations.”
One of the advantages of Datumo’s evaluation tool is that it can automatically generate test cases. “We automatically generate the questions to ask AI models—the test sets for the AI evaluation,” says Michael Hwang, cofounder and chief strategy officer of Datumo. “We use our own AI system to automatically build the test cases, perform the evaluation and analyze the evaluation.”
Datumo also provides other AI development services such as data for pre-training foundation models. For example, the startup uses optical character recognition (OCR) to convert printed text from books into training datasets.
Earlier this month, a consortium that included Datumo, SK Telecom, Krafton and AI chip startup Rebellions was among the five consortia selected by South Korea’s Ministry of Science and ICT to develop a proprietary AI foundation model. Datumo is helping the consortium with AI data engineering.
Datumo is currently expanding in Japan and the U.S. Hwang relocated from Seoul to Silicon Valley to lead the global expansion.
Datumo is the latest AI-related startup in the semiconductor hub of South Korea to raise funding. In March, Movin, a startup that develops AI-based real-time motion capture devices, raised $2.7 million in a pre-Series A funding round from the likes of Atinum and Naver D2SF. Superb AI, also led by a Forbes 30 Under 30 Asia alum, raised $10.2 million in a Series C funding led by Doosan Investments in October.
More from Forbes