Company File / Computer Vision Infrastructure
Superb AI.
The company that decided the least glamorous job in AI - labeling the data - was worth building a business around.
The company that decided the least glamorous job in AI - labeling the data - was worth building a business around.
Here is a fact about artificial intelligence that does not fit on a keynote slide: before a computer can learn to see a pedestrian, a cracked weld, or a ripe strawberry, a human being has to sit down and draw a box around thousands of them. This is called data labeling. It is tedious, it is expensive, and it is the single most common reason a promising AI model quietly fails in production. Superb AI, founded in 2018, looked at that unloved chore and decided it was, in fact, the business.
The pitch is refreshingly free of magic. Superb AI does not claim a secret algorithm that will make your model brilliant. It claims something narrower and more useful: that it can get you from a pile of raw images, video, and 3D LiDAR scans to a working, deployed model faster than you can currently manage. The company calls the destination "the AI you want, faster than ever." What it is really selling is the removal of friction - the month your machine-learning team would otherwise lose to babysitting a dataset.
It is a Korean-American company, which is to say it lives in two places at once: a headquarters in San Mateo, California, and deep roots in Seoul, where much of its customer base and engineering gravity sits. It went through Y Combinator's Winter 2019 batch, which is the sort of credential that opens Silicon Valley doors, and then it went and raised most of its later money from the pillars of Korean industry - Hyundai, Samsung, Doosan, Kakao - which tells you where it actually makes its living.
"Vision AI that serves as the eyes of industry."
Superb AI, on what it buildsThe founder is Hyun Kim, who holds a robotics PhD from Duke and did a stint researching machine learning at SK Telecom's T-Brain lab before landing on a list of Forbes' 30 Under 30. He started the company with four co-founders - Moonsu Cha, Jung Kwon Lee, Jonghyuk Lee, and Hyundong Lee - a founding team heavier on research pedigree than most. That heritage shows in the product: this is a company that treats data quality as an engineering discipline rather than a staffing problem you throw offshore contractors at.
The result is a platform, not a service. And the distinction matters, because it is the difference between renting people to draw boxes and selling software that helps machines draw most of the boxes themselves - then flags the ones that got drawn wrong.
The Superb Platform (once called Superb AI Suite) covers the whole life of a computer vision model. You can run it in the cloud or, for the factories and government agencies that cannot let their data leave the building, entirely on-premise. Here is what each part actually does for you.
AI-assisted labeling that auto-annotates large image and video sets, then uses generative AI to fill the gaps where real examples run thin. You spend your time reviewing, not clicking.
Automated dataset curation built on embeddings - the AI compares images the way a model "sees" them, surfacing near-duplicates, mislabels, edge cases and imbalances so your training set is balanced and representative.
Model training, diagnostics and deployment wired directly into Label and Curate, so going from cleaned data to a running model happens without messy hand-offs between tools.
The deployment and integration layer that puts a trained vision model into a real production workflow - on the cloud, or out at the edge on hardware like NVIDIA Jetson.
"Curate auto-builds datasets that are balanced, representative and optimal for model performance - and surfaces the edge cases, mislabels and imbalances humans miss."
How the curation engine worksMore than 100 enterprises run on the platform, clustered in the industries where a camera making the wrong call has real consequences: autonomous driving, smart factories, security and surveillance, and the public sector. When your defect detector misses a crack in a steel beam, the problem is almost never the algorithm - it is the data. These are the companies that figured that out.
The interesting thing about Superb AI's cap table is not the amount - it is the names. A data-labeling company that pulls strategic checks from Hyundai, Samsung, Doosan and Kakao is a company those firms need for their own AI ambitions. In the first half of 2024, Superb AI said it had already matched all of the previous year's revenue, and set a target of more than 200% annual growth.
Bar length reflects round size. Series C investors: Doosan Investment, Hyundai Motor Group, Kakao Investment, Samsung Next, plus returning backers KT Investment and Premier Partners.
Hyun Kim and four co-founders launch the company to automate the preparation of computer vision training data.
Joins YC's Winter 2019 batch and raises roughly $2.1M, launching the Suite platform.
Premier Partners leads a round to scale the training-data platform.
Korea Development Bank leads; Superb AI ships embedding-powered automated curation.
Superb Model and Apps extend the platform from labeling all the way to training and deployment.
Doosan, Hyundai, Kakao and Samsung Next invest; cumulative funding reaches $36.9M as revenue roughly doubles.
Co-founder and CEO, robotics PhD from Duke, former machine-learning research engineer at SK Telecom's T-Brain, and a Forbes 30 Under 30 Asia honoree. He is the kind of founder who talks about data quality the way a manufacturing engineer talks about tolerances - as something you measure, not something you hope for.
"To transform every industry with AI, fueled by state-of-the-art tech, superb products and services."
The company missionSee the platform in motion and hear the team explain the thinking behind it.
It provides an end-to-end computer vision MLOps platform that lets teams label, curate, train, deploy and monitor AI models on images, video and 3D LiDAR data.
Founded in 2018 by Hyun Kim (CEO) with co-founders Moonsu Cha, Jung Kwon Lee, Jonghyuk Lee and Hyundong Lee. It graduated from Y Combinator's Winter 2019 batch.
About $36.9M cumulatively - a seed round, a $9.3M Series A (2021), a $16M Series B (2022) and a $10.2M Series C (2024).
Over 100 enterprises across autonomous driving, manufacturing, security and public sector, including Samsung, LG, Qualcomm, Hyundai Motor, Toyota and Nippon Steel.
It combines AI-assisted labeling, embedding-based automated curation, generative synthetic data and integrated model training and deployment in one platform - available in both cloud and on-premise forms.
Headquarters: 400 Concar Dr, San Mateo, California 94402 · +1 650-495-4104 · and an office in Seoul, South Korea.