Datumo

Datumo is a South Korean AI company whose main product, Datumo Eval, is a platform for evaluating large language models and LLM-powered applications. Users upload source documents and the system uses an agentic flow of multiple cooperating LLM agents to generate industry-specific "golden" question sets, then scores model answers against default or custom metrics. The platform also assesses the quality of the generated questions themselves and includes human-alignment mechanisms intended to reduce the gap between the evaluator's intent and automated results. The interface is designed to require no coding, targeting non-technical functions such as policy, compliance and trust-and-safety teams, and it automatically produces test data and evaluations to surface unsafe, biased or incorrect model outputs.

Alongside evaluation, Datumo offers an AI Red Team service that probes generative AI models for vulnerabilities that internal testing may miss, iteratively adapting attack strategies and prompts. Its published risk taxonomy covers personal information disclosure, biased content, misinformation, cybersecurity threats, inappropriate advice and dangerous content such as material related to weapons or self-harm. The company also produces evaluation datasets, including KorNAT, described as the first LLM evaluation dataset on Korean social values and common knowledge, whose paper was published at ACL 2024 with Datumo staff as first and third authors.

The company was formerly known as SelectStar and began as an AI data-labeling business before moving into model evaluation as clients asked it to compare and assess AI outputs.

Founding story

Press coverage reports the company was founded in 2018 by five KAIST alumni. CEO David Kim, previously an AI researcher at Korea's Agency for Defense Development, started the company after encountering inefficiencies in traditional data labeling and built a reward-based app letting people label data in their spare time for pay. During a KAIST startup competition the team secured pre-contract sales worth tens of thousands of dollars from alumni-led startups and other businesses before the app was finished.

Business model

Datumo sells AI evaluation and red-teaming services to enterprises, positioning its Datumo Eval platform for teams deploying LLM-based products across e-commerce, finance, education, legal, healthcare and customer service. The website routes prospective buyers through demo requests and a "talk to sales" path, indicating a direct enterprise sales motion.

Enterprise sales of the Datumo Eval platform and evaluation/red-teaming services; the company reported approximately $6 million in revenue in 2024 and more than $1 million in its first year of operation.

Traction

Reported approximately $6 million in revenue for 2024, more than 300 clients in South Korea, and 150 employees based in Seoul. Early enterprise customers cited include Samsung, Samsung SDS, LG Electronics, LG CNS, Hyundai, Naver and SK Telecom. Total capital raised is reported at roughly $28 million following the August 2025 round.

Latest developments

In August 2025 Datumo announced a $15.5 million round led by Salesforce Ventures with KB Investment, ACVC Partners and SBI Investment participating, bringing total funding to about $28 million; the raise followed an eight-month process that began after Salesforce Ventures saw a LinkedIn post about a fireside chat CEO David Kim held with Andrew Ng. Proceeds are directed to R&D in enterprise AI evaluation and to expansion in South Korea, Japan and the U.S. The company continues to publish product and research posts, including material on LLM version management and an AI Red Team Challenge tied to MWC 2026.