The team we will meet today is the winner of 2021 Spring E*5 KAIST, GSD. GSD is a team that generates virtual data for AI training based on digital twin technology.

Let’s meet CEO Jaewoon Kim of GSD!

Hello, I am Jaewoon Kim, CEO of GSD. GSD is a technology startup team of KAIST master’s and Ph.D. students developing AUTODATA, a virtual data generation solution service. We endeavor to solve the most serious problem in AI development pipeline of data collection and processing.

Virtual data generation solution might sound unfamiliar, but AUTODATA is explained in detail below.

Many people are curious about the meaning of GSD. Some say it stands for ‘Great Super Data’ and others know it as ‘GunSsakDol’. However, GSD actually stands for ‘Gabsida!’ (Let’s Go!) ???? It was sort of a joke, but it has a deep meaning. When GSD was first established, the team members gathered at the Eoeun-dong café ‘Gabe’ and then the Japanese restaurant ‘Yoshida’. We decided to name the company ‘Gabsida’ to remember our initial resolution and to keep going on to the end.

It’s special to name the company after the place where the team members decided to embark on a journey.

There were many difficulties until choosing the current startup item. Since August 2020 when the team was first formed to winning E*5, our item changed 4 times. Rather than gathering to solve a specific problem, the team gathered in desire of starting a business so we went through hardships.

In search of various items, we discovered a problem that directly relates to our current item thanks to the Myungchul Kim crew. While I was working on an AI related project at the laboratory, the cost of building a dataset for training was beyond my expectation. The reason behind this was because the task was labor-intensive because workers had to create the dataset manually. This made no sense to me. AI exists to make people convenient, but for the AI to function well, it requires the hard labor of many people. I discovered the problems in data creation in terms of time and the consistency and diversity of data. Among the numerous problems, creating 3D images or Point Cloud data is extremely difficult and time & cost-consuming. I realized that data collection and the high level of processing hinder the development of AI, so decided to develop AUTODATA which could solve these issues innovatively.

AUTODATA is a digital twin-based virtual data generation service for computer vision AI training. Data is a crucial component in the development of artificial intelligence in that data collection and processing take about 80% in the AI development pipeline. However, these tasks, especially those that involve image data, have been handled manually by people through repetitive and simple labor, thus resulting in the high time costs and low accuracy. AUTODATA is a solution that creates a virtual environment and then create and augment data to solve the aforementioned issues. Specifically, AUTODATA generates image data in the 3D simulation environment with the same settings as reality then changes conditions such as illumination, shadows, and textures to extract training data needed in training. The digital twin data made during the process is saved digitally, so it does not require additional processing by humans. Currently, we provide training data needed in the AI robotics market.

Data plays an important role in the AI development process, but it is surprising that the data processing tasks have been continuing with human labor. It seems like the AI field still has a long way to go. Using AUTODATA developed by GSD during the data creation and processing can save time and money without additional human labor.

Existing AI data processing solutions are based on manual human labor. Methods like crowdsourcing which is relatively cheap and fast, but these are unsuitable for tasks that have high processing difficulty like 3D image data. We create and augment data virtually to provide datasets for AI models at 10,000 times lower price than existing solutions, and 120 times faster speed. Since we create data digitally, we can produce a 100?curate high-quality data without additional labor. This helps customers operate their AI development pipelines at a much faster and improved performance.

There are more and more startups that process data, but there seems to be a slight difference in the methods. Creating data in a virtual environment will be able to deliver datasets to customers at a much faster and cheaper rate and establish AI development pipelines quickly. It seems like more customers will look forward to GSD’s AUTODATA.

Our team has been friends for a long time. We were all very interested in startups, and in August 2020, crew Kang Hong-gu contacted me to work on a small project together. At first, he and I worked on the project, then I contacted crew Kim Myung-chul to create GSD today ????

The reason why we became interested in startups was because we all had a dream of impacting the world beneficially. In my case, I believed that my work inspires people to think and act in a different way what they previously took for granted. I always dreamed of changing the world with my work, and I chose to become an entrepreneur to realize this dream.

It’s not easy to challenge starting a business with the desire to change the world beneficially, but CEO Kim has realized this dream. Science and technology are able to develop with the work of teams like GSD.

E*5 KAIST was like a booster. Before participating in E*5 KAIST, our team prepared to start a business, but since it was our first time, we worried whether we were taking the right path. E*5 is a grateful program that allowed us to grow exponentially when we were in need of quick execution and feedback.

The biggest lesson I learned while participating in E*5 was the persistent mindset to discover customer needs and problems in the market, and the growth through repeated trials and feedback. The program to closely verify ideas and the detailed advice from mentors at the nation’s top venture capital and accelerators made the GSD today. I want to take this opportunity to thank all mentors and E*5 program staff members for helping us ????

Startup support programs like E*5 KAIST can be helpful when preparing to start a business. Experts become mentors to provide realistic advice and solution based on marketability, development potential, and business feasibility, so as CEO Kim said, this program can become a booster. We hope many people stay interested and participate in Startup KAIST’s startup support program!

Our goal is to establish an automation pipeline technology that provides image data that clients need at all times within a year and a half, and enter the global market with this. Unlike the relatively small domestic market, virtual data services are a new trend in the global infrastructure market for machine learning service operation (MLOps). The biggest challenge is to quickly enter this market and grow into a global company.

An automation pipeline technology that provides data at all times and collaboration with other AI companies will help GSD enter the global market and become a global company.

Our vision is “Dream to Real with AI”. What will the role of humans be in 30 years? The answer we thought of was “keep dreaming”. Even if all other things are automated, thinking of new ideas and challenging lie in the role of humans. Then what will the role of AI be? We believe artificial intelligence will become companion who will help human’s dreams come true. To realize this, we will develop a virtual data generation service through self-supervised AI. Through this, we plan to advance AI development by 10 years.

In the process of creating and processing data in artificial intelligence development, GSD produces and extract virtual data. This allows timely and cheaper development, and enhances AI technology. Although AI has developed a lot compared to the past, AI technology will develop even more because of teams like GSD.

We look forward to and support the development of GSD. This was CEO Jaewoon Kim of GSD.