Seoul-based Datumo raises $15.5M to tackle Scale AI, backed by Salesforce | TechCrunch

Most organizations say they aren’t absolutely ready to make use of generative AI in a secure and accountable means, in response to a current McKinsey report. One concern is explainability — understanding how and why AI makes sure selections. Whereas 40% of respondents view it as a major danger, solely 17% are actively addressing it, per the report.

Seoul-based Datumo started as an AI knowledge labeling firm and now desires to assist companies construct safer AI with instruments and knowledge that allow testing, monitoring, and enhancing their fashions — with out requiring technical experience. On Monday the startup raised $15.5 million, which brings its whole raised to roughly $28 million, from buyers together with Salesforce Ventures, KB Funding, ACVC Companions and SBI Funding, amongst others.

David Kim, CEO of Datumo and a former AI researcher at Korea’s Company for Defence Growth, was annoyed by the time-consuming nature of information labeling so he got here up with a brand new concept: a reward-based app that lets anybody label knowledge of their spare time and earn cash. The startup validated the thought at a startup competitors at KAIST (Korea Superior Institute of Science and Expertise). Kim co-founded Datumo, previously referred to as SelectStar, alongside 5 KAIST alumni in 2018.

Even earlier than the app was absolutely constructed, Datumo secured tens of hundreds of {dollars} in pre-contract gross sales through the buyer discovery part of the competitors, largely from KAIST alumni-led companies and startups.

In its first yr, the startup surpassed $1 million in income and secured a number of key contracts. At present, the startup counts main Korean corporations like Samsung, Samsung SDS, LG Electronics, LG CNS, Hyundai, Naver, and Seoul-based telecom large SK Telecom amongst its shoppers. A number of years in the past, nevertheless, shoppers started asking the corporate to transcend easy knowledge labeling. The seven-year-old startup now has greater than 300 shoppers in South Korea and generated about $6 million in income in 2024.

“They wanted us to score their AI model outputs or compare them to other outputs,” Michael Hwang, co-founder of Datumo, instructed TechCrunch. “That’s when we realized: we were already doing AI model evaluation — without even knowing it.” Datumo doubled down on this space and launched Korea’s first benchmark dataset targeted on AI belief and security, Hwang added.

“We started in data annotation, then expanded into pretraining datasets and evaluation as the LLM ecosystem matured,” Kim instructed TechCrunch.

Meta’s current $14.3 billion acquisition-like funding in data-labeling firm Scale AI highlights the significance of this market. Shortly after that deal, AI mannequin maker and Meta competitor OpenAI stopped utilizing Scale AI’s providers. The Meta deal additionally alerts that competitors for AI coaching knowledge is intensifying.

Datumo shares some similarities with corporations like Scale AI in pretraining dataset provisioning, and with Galileo and Arize AI in AI analysis and monitoring. Nonetheless, it differentiates itself by its licensed datasets, significantly knowledge crawled from revealed books, which the corporate says gives wealthy structured human reasoning however is notoriously troublesome to wash, in response to CEO Kim.

Not like its friends, Datumo additionally gives a full-stack analysis platform referred to as Datumo Eval, which routinely generates take a look at knowledge and evaluations to examine for unsafe, biased or incorrect responses with out the necessity for guide scripting, Kim added. The signature product is a no-code analysis instrument designed for non-developers like these on coverage, belief and security, and compliance groups.

When requested about attracting buyers like Salesforce Ventures, Kim defined that the startup had beforehand hosted a fireplace chat with Andrew Ng, founding father of DeepLearning.AI, at an occasion in South Korea. After the occasion, Kim shared the session on LinkedIn, which caught the eye of Salesforce Ventures. Following a number of conferences and Zoom calls, the buyers prolonged a tender dedication. Your complete funding course of took about eight months, Hwang mentioned.

The brand new funding might be used to speed up R&D efforts, significantly in growing automated analysis instruments for enterprise AI, and to scale world go-to-market operations throughout South Korea, Japan, and the U.S. The startup, which has 150 workers in Seoul, additionally established a presence in Silicon Valley in March.

We’re at all times trying to evolve, and by offering some perception into your perspective and suggestions into TechCrunch and our protection and occasions, you’ll be able to assist us! Fill out this survey to tell us how we’re doing and get the possibility to win a prize in return!