Cha Moon Soo, CTO of SuperbAI, “We will evolve from a one-stop solution to an active AI… and implement ‘Agentic Vision AI’”

 

“Multimodal AI, vertical AI, and agents may seem like distinct technologies, but they are, in fact, part of one ‘connective body.’ Moreover, companies now seek ‘proactive AI,’ and SuperbAI is advancing to address this demand.”

CTO Cha Moon Soo cited the company’s “one-stop platform” as the reason why SuperbAI has been at the forefront of the domestic industrial Vision AI field since its establishment in 2018

Originally, the company was famous in Korea and abroad for data processing, or labelling, in its early days. However, it found that the companies it encountered in the field had different needs, manpower, and expertise, and were having difficulty with technology. Therefore, it came up with a platform that could handle everything from data processing to AI model building, deployment, and operation. This is the ‘Superb Platform’

This enabled them to apply Vision AI to Korea as soon as possible. This was also the case in Japan, where a Toyota official who adopted the Superb platform in May last year said, “There is no platform in the Japanese market that has all the ML ops functions yet, so it is receiving a good response locally.”

CTO Cha Moon Soo stated, “The reason we have been able to evolve into the current 'all-in-one platform for Vision AI adoption' is that we have continuously developed technology one step ahead, based on the needs and communication with our customers.”

He also added that based on on-site demand, the company is now advancing its technology to the next level. He stated, “As AI is being actively adopted and leading to tangible improvements in work efficiency, companies are now seeking more ‘proactive AI’”

Proactive AI goes one step further from the 'input-AI-output' structure and provides something after the 'output'.

He cited security and monitoring as examples. With the recent strengthening of laws related to serious industrial accidents, 'situation analysis and insight extraction' has become an essential element, going beyond simple object detection by cameras.

AI can generate statistics indicating, for example, that the incident rate increases at specific locations or times. Delivering such information, which is crucial for proactive measures, prevention, and reducing human casualties, has become increasingly important. SuperbAI's technological goal is not to 'replace humans' but to complement areas where humans cannot operate effectively.

“For active AI, we need to have a structure of multimodal-vertical AI-agents,” said CTO Cha.

Multimodal, vertical AI, and agents are hot topics in the AI industry these days, but they are often perceived as individual technologies, he noted. ‘For agentic vision AI, the three parts must be connected in a coherent way,’ he said.

First of all, multimodal is important because data is generated in various parts.

In order to properly build a vision AI, it is necessary to process various data such as sensor values, text, and voice, as well as images and videos through CCTV.

Autonomous driving is considered to be an area where this diversity of data is maximized. There is an infinite amount of data to be collected, including cameras, radar, LiDAR, voice, and logs. Robots also come in various forms, from sensor values to location information, and multimodal functions are essential for reading them. 

To explain vertical AI, which refers to expertise in a specific domain, the example of a medical CT scan was used. The amount of information that can be obtained by the average person from a single image is completely different from that of an expert. In other words, information can be seen as much as one thinks and finds. Vertical AI is essential for applying this to each field.

Finally, the agent acts as a system that encompasses all of these. Recently, Google defined the structure of the agent as “goal setting, environment observation, and self-judgment.” In other words, it means that AI should be able to derive insights based on the data it collects without having to inject a goal every time.

Specifically, it introduced that it has worked on the following: ▲ foundation model of 'object detection' based on zero-shot ▲ video analytics solution ▲ advancement of 'digital twin' of camera-based 3D spatial mapping.

The zero-shot model is intended to respond smoothly to dynamically changing on-site conditions. For example, in the logistics field, existing vision AI often failed to recognize parts that were slightly different in appearance or size. It is no wonder that this took a lot of time and money.

Therefore, SuperbAI is developing a more advanced foundation detection model, a type of vision language model(VLM) that can detect objects based on zero-shot. It is a model that adds visual information to language concepts and can match existing concepts when faced with new objects. In other words, it is a zero-shot model that can learn on its own.

“However, even if you have all of these features, if you rely on only some of the cameras, information management can be limited,” said CTO Cha. ‘In the end, vision AI has no choice but to move towards 3D spatial mapping, or digital twins.’ He said that this allows you to structure the site and derive insights.

The company is eyeing the defense industry as a next-generation business area. The reason why SuperbAI is stepping up its deployment of edge AI is because demand is increasing in the defense, manufacturing, and logistics industries.

In particular, since it is difficult to use the cloud for security reasons in the military, he added that he is discussing collaboration with a domestic NPU company to strengthen edge capabilities.

He also emphasized that SuperbAI has the advantage of being able to continuously update or respond immediately, saying, “It is common for most vision AI models to be updated every one to two years after hardware application.”

Finally, he emphasized, 'AI is no longer a future technology—it is time to demonstrate its ability to generate revenue.' He added, 'This year, we will focus on proving that AI can help companies reduce costs and increase profits.' SuperbAI is also aiming for an initial public offering (IPO) in 2026.

https://www.aitimes.com/news/articleView.html?idxno=168011