Day1_트랙파트너사

Shared on August 21, 2026

[0:00] I need to do it.

[0:09] So we will now begin the track announcement session. [0:14] This track partner will introduce their track. [0:18] Remember that you may choose only one track. [0:22] If you have any questions, please visit your track partner's booth while the partners are available. [0:29] If no one is at the booth, please ask your question in the Discord channel.

[0:35] First, please welcome Gyeongsang-guk-do, Post-Tek, and Microsoft. [0:43] Let's give a big offer for you.

[1:08] I'm honored and pleased to speak in front of all these young and talented people here. [1:12] Well, I guess all of you guys are super passionate about this Jungshun X Korea 2026. [1:19] My name is Ryan and I work as an AI business solution specialist for Microsoft Korea. [1:26] Today I become very honored to speak on behalf of Microsoft Jungshun-buk-do province and POSTEK to present you guys a track challenge for Jungshun X Korea 2026. [1:32] And the panel of today's track for our team is Unlocking Local Innovation Through Public Data. [2:12] This means the public data is already open and more than welcome to be utilized. [2:19] The main challenge is to make that data available for AI to use it digitally. [2:25] While there have been already a lot of advocates and movements made among the public, some people have already started. [2:35] Today, I’ll give you three examples where an individual or a project team have used public data to make the world better by using AI and data infrastructure. [2:42] Well, the first example was done by a single talented developer in Korea. [2:46] A developer named Junghung Park has migrated entire Korean legislation and regulations, including the limited history of 80,000 carriers in a single line into a state depository by using the Markham method.

[3:07] And due to this effort, the entire public are now accessible to the Korean legislation and regulation into open source where the developers can use AI to process this data set to make the law process and search of the law into a lot more streamlined and efficient way. [3:29] Now, people are able to use this dataset to build an agent or put it into the data lake [3:39] in order to make something out of the new product. [3:45] And the second project was done by a Korean government officer. [3:52] And this project was to make the Korean legislation and law cases into an MCP connector. [3:59] Due to this project, now people are able to use the Korea legislation and all the track historian law pieces into an NCP connector and simply build an AI agent out there. [4:06] The public official wrapped the Ministry of Government legislation data on an NCP server and released as an open source, which gave a huge impact to the Korea legislation society and community. [4:25] Now a lot of lawyers, government officials, are now using the ZSN to make the world better and also to make the law a lot more accessible and easier to the public.

[4:36] And third, it's a little bit more friendly and a lot more open for the general community. This can be applied to the third sample I've given. [4:44] One single developer has actually processed the real-time location data of the buildings and facilities in South Korea to build an app where it can guide a very general public to find the shade zones when the weather is too hot and during the hot times. [5:03] And this was led by a single university student. [5:07] Now, it's your time. [5:12] It wasn't the system or the problem for any of the other people that moved all over the top of the golfers to come up with the same idea. [5:20] But the cool of one person that moved first was one in England.

[2:35] Today, I’ll give you three examples where an individual or a project team have used public data to make the world better by using AI and data infrastructure. [2:42] Well, the first example was done by a single talented developer in Korea. [2:46] A developer named Junghung Park has migrated entire Korean legislation and regulations, including the limited history of 80,000 carriers in a single line into a state depository by using the Markham method.

[6:15] And this applies to the first and second example that I gave. The constraint is that at least one public data piece is required. If you choose to build infrastructure, you must specify your open source license and publish it on GitHub. It should be ready to run. What does that mean? The solution should be very straightforward, easy, and even be utilized by any non-conventional data. Third, the open source contribution. Teams working on data infrastructure projects must publish their code on GitHub and clearly specify the license. And those data sets that we believe is worth and also we believe is beneficial for the profits could actually be officially utilized and applied by the government of the profits for the related developers.

[8:34] Fourth and last, the prototype bonus is there.

We will give additional points to the tangible deliverables such as chatbots, web services, mobile apps, or live demos. Again, we expect all of you guys to create impact, the real impact. And please, please avoid just using simple user interface without actual data integration or exposing the original data without any processing or putting into the data foundation. Our judging criteria is based on five different methods. It's weighted and evaluated. The first criteria, of course, is in technical excellence. This is by far almost the most important thing. The solution quality, technical soundness, and effective use of AI and public data will be weighted 25%. It should be deliverable. The valuation criteria. We'll see how deeply the team leverages Office Studio and its APIs. It goes for 30 points. And service should be completed. It should run, hopefully, flawlessly. And credit fee always matters. And product planning. You should find your market bid. Concrete hard to users and follow the definition. So these are the scoring bases. If you make your own score, you can download your score template from the JSON file. And you can just copy and paste into our site. And here you can score the performance of your score compilation.

[9:26] And the reason why this is weighted 25% instead of 35% is that we value the public worth and the public ability. The potential to address real public needs and create measurable local impact will be weighted 35%, which is the biggest portion of the project.

The whole point of having this hackathon is to come up with the innovative idea and make it into a real product and services. Originality, creativity, uniqueness, and comparison to existing approaches will be evaluated. Sustainability and sociability are also very important. We would like to see long-term viability, maintainability, and potential for the QRR object. This criterion is rated 15%. Last but not least, we will give a prototype bonus. Additional credit for working demonstrations such as check-ups, web apps, dashboards, or circular-related services will be provided with an extra bonus of 10%.

[8:34] Fourth and last, the prototype bonus is there.

[6:49] You're more than welcome to freely discovering and analyzing various regional issues in terms of the province, using open data and AI, and developing AI-based services and solutions to solve regional problems.

[6:20] Public data related to Gyeongsangbuk-do Province or those helpful to Gyeongsangbuk-do administration is processed and disclosed in a format that AI can immediately use, such as MCP server, open source skills, and plugins in a standardized manner.

[6:36] Or the second way you can do it is to come up with the regional problem solving services.

제출물은 다음 조건을 충족해야 합니다. 단순히 처리하거나 공개만 하지 마세요. 실제로 업무를 수행할 수 있는 프로토타입도 제출해 주세요.

[37:16] your family. So we provide some pricing and things for you. We already know about that. [37:25] So also if you become the leader, we will provide the company for you. Even though you don't like, we will invite you to our company.

[8:08] Third, the open source contribution. Teams working on data infrastructure projects must publish their code on GitHub and clearly specify the license. And those data sets that we believe is worth and also we believe is beneficial for the profits could actually be officially utilized and applied by the government of the profits for the related developers.

[8:34] Fourth and last, the prototype bonus is there.

[11:45] thank you Ryan. so we're going to next track park. so next please welcome upstage. give a big uppost here.

[38:30] Please choose your track campaign. You have until midnight to submit your kickoff mission proposal. And also the level up and AI will give you their own session at 9:00 PM about their challenge and about their platform.

[12:22] Today, I'm going to give you a quick introduction to Offstage and have the challenges we prepare for. Offstage is a global AI startup. We have headquarters at Korea, US, and Japan. We have a mission of making AI big a vision and vision of building intelligence for the future of work. [12:44] We've been raised 270 million dollars so far and we acquired the well-known portal down as of this year. [12:53] We are Korea's first generative AI mini-corp. [12:58] Some of you may know the Ministry of Science and ICT announced the sovereign AI foundation models this week. [13:07] We are honored to rank top three, along with SKT and LG.

[13:13] We've been leading the AI market and this journey started from January 2024 by launching Solar Mini. [13:23] Since then, we've been expanding our foundation models to document parts and Solar Pro. [13:30] This year, we are focusing more on agents and that is why we built our platform, Upseat Studio, which is the unique audience for this hacking. [13:41] So, Clulop 04 is specialized in HMQ spaces.

[14:12] So people know Solar, but not many people know we are traditionally OCR companies. [14:21] In 2023, we launched our first cycle, Document OCR, and by far we've been doing great. [14:30] And me myself uploaded Document Intelligence models on the AWS Marketplace, and I'll have to see they are on the second place.

[13:46] So we have two foundation models, LLM and document parse. [13:52] And we also have platforms that realize those two foundation models, Upstate Studio. [13:58] And most importantly, we have Upstate Education Team, our most valuable asset. [14:03] And thanks for that, they are always at the forefront of the AI era and maybe the Korean AI. [14:10] So huge applause for them.

[14:39] We have many partners and customers in Korea and outside of Korea. [14:47] And among them, Hanan Life in 2002, they were our first customers. [14:53] We signed a contract even before we made our product. [14:57] There are lots of many creative things. [14:59] So if you are curious about Korean SI or how to study,

[15:07] There are a lot of behind stories. So why do they work with Outstage? Why do they work with us? I want to point out two things. So our bottom-section nodes, axons and benchmarks, some of them said Outstage was only good at benchmarks. But for a lot of people, it's actually good. Now they're starting to use them.

[15:30] And as I said, we are going to be for top three by administration of science and ICT. [15:37] So we'll see who wins. [15:39] And secondly, scalability. [15:42] So we are accessing over that 3 million of pages every day. [15:46] I was at the upstage LLM. [15:49] I like that. [15:51] Because I was a good guy. [15:53] I was a good guy. [15:55] And I was a good guy. [15:57] I was a good guy. [15:59] I didn't know. [16:01] I was a good guy. [16:03] I was a good guy. [16:05] I was a good guy. [16:07] I was a good guy. [16:09] I don't know.

[12:11] Hello, Dumb-Tip.

That's possible because we have core agents, classified parts, extract and LLM. The first feature is Chat. Once you download a document, you can check, search, translate. For example, when you download a receipt, you can ask what the format is or what the name of the floor. Think of it as a receipt. When you wrote a receipt, you can extract the total amount, the name of the stores, the date you have issued, etc. And I don't want to use it. I don't want to use it. I can use it. DocMint has lots of features such as boxy, multi-pitch tables, complex tables. We support basically all the DocMint features. It has a limit of $1000 pages for recasts. Speed and accuracy shouldn't be a problem. And you can set the schemas. It can be also generated automatically. You can set the instructs. The third parsing result is classified as "공공인우에". It extracts all the key information like prices.

Like papers about medicines, you can tell it's good or qualified. It's classified as "gine bonante". Features are extracted. And this is not applicable because of that.

그리고 인스트럭스에 공급 지원이 가능하다. 그래서 유용한 유지 케이스가 있으면 이렇게 만들어 볼 수도 있고. 근데 이거 쓰고 싶다. 저거 편하겠다. 나도 좀 욕심 나요. 근데 업스테이지로 하면 어쩔 수 없이 무조건 일하는 생산성으로 가야 될 거고.

It is about Ministry of Good Negros. 저런 거 아니에요? 저런 LH 이런 거 아니에요?

And you can see the sources as well.

[21:52] Next, we introduce you to a track challenge.

[21:59] Simply put, automate paperwork using Opsi Studio. But one repetitive task requires a mistake for everyday hassle caused by documents or an issue, and beyond the document age of service. You should use Opsi Studio and integrate with any service PRS APIs and records. Most importantly, Opsi Studio must power the core document processing stages.

So your choice is to make a commercial reaction to solve your hard problems. But it should be a third-general because we will provide you some challenging topics.

[34:42] automatically unscored by our system. [34:46] Also, we recommend you to provide interactive visualization with how agents are communicating with each other, or how agents are working, and how many agents are interacting or joining together to solve similar problems.

[21:47] We have tutorials you can easily follow.

[21:44] If you have any tracks, you can follow them.

[21:39] This is the link for the studio.

[21:33] This one is also not available.

[21:30] This one is here to do it.

[23:26] So you can use Upstate Studio's own website and once you complete the work flow, you can call it as an API. [23:37] If you have any problems or passions, you can find it there. [23:39] So any questions are always welcome. [23:45] Thank you for listening to my presentation.

[23:15] How to use Optage APIs? You can sign up for Optage console and read it code that you will receive free credits.

[23:01] Prizes and benefits will give you goody packages and $50 worth of credits and for Winnie to members, $200 worth of credits.

[23:15] How to use Optage APIs? You can sign up for Optage console and read it code that you will receive free credits.

[23:01] Prizes and benefits will give you goody packages and $50 worth of credits and for Winnie to members, $200 worth of credits.

[23:15] How to use Optage APIs? You can sign up for Optage console and read it code that you will receive free credits.

[24:28] So it's very hard to speak in English, so I'm really happy to speak English.

[24:26] I'm so excited to talk to you.

[24:25] She's nice.

[24:21] Sorry, my English tokenizer is broken.

[24:20] Oh, I'm sorry.

[24:13] Hi, everyone.

[24:07] Next, please welcome Level Up and to your host, I'll give a big applause to you.

[24:05] I'll send you a link.

[23:59] I'll send it to the link.

[24:33] So let's introduce today's charity with Levera and Pyrrhoz and Levera. [24:41] So who is Levera? [24:43] We are Levera. We want to upgrade the Levera to the world. [24:49] So we found Levera in 2015 with the name "Levera" type right now.

[24:55] So we want to democratize the AI strategy. We call it intelligence for humanity. [25:03] So today you will enjoy the intelligence of your PC with your AI, CPU, power, and others. [25:13] So we are working toward a world where intelligence grows like electricity. [25:18] Because you already know about the phenomenon. Because we are always using AI knowledge. [25:26] So it's inevitable to be accustomed to the AI and enjoy the AI and work and play with AI.

[25:38] So we are making an accessible for 11 years. [25:44] So now we are making AI scalable and so we have a health potential and we have a growth of the brand. [25:58] So we fund our company with open source intelligence and open source contributors. [26:04] So we have many many members, open source contributor members in our company, Okro2Me. [26:10] So we are also the sponsor of the Cloud Native Community Foundation, Lightwatch, Pryson, AI and the Wings Foundation too. [26:19] So we have a Wings Corner contributor, Pryson and Nephi and many many countries. [26:27] So we are very good and enjoying the development process.

[26:48] We are making AI infrastructure platform. Nowadays, more and more

[26:42] There were no customers at the time. Nobody was interested in our business.

[26:34] We found our company 11 years ago. We had a hard time for almost 5 or 6 years.

[26:34] We found our company 11 years ago. We had a hard time for almost 5 or 6 years. [26:42] There were no customers at the time. Nobody was interested in our business. [26:48] We are making AI infrastructure platform. Nowadays, more and more

[27:03] through the world. [27:04] So we are powering many customers, especially in South Korea and among five more countries in the world. [27:13] And also, if you are using a Galaxy smartphone, you are already using it on the second-year.

[27:23] So we have a proud customer, company Tonics, or Egeo Tonics, and other companies who are making AI. [27:35] So we are very proud about it. [27:38] Because many project models and other users built in South Korea are built on Backdoor AI. [27:46] So if you are living in South Korea, you are always using or experiencing the AI models built on Backdoor AI.

[27:56] So we are looking at the over-mini-peak AI over in G-SAN. [28:05] So we want to combine the hyper-computing and AI to be a single platform. [28:12] Because these two areas, the field are aiming very different. [28:19] So we are now using the high performance computing and AI to the Cinecter platform. [28:27] And also we are making the heterogeneous AI hardware platform to be accessible and usable. [28:36] Because nowadays more and more people are using very complex hardware including the Paragon from NVIDIA and also the Pures AI and the other AI-gencriptors that I think Pures is the best reach of the center.

you will enjoy the period the AI is powered today. So we are making that kind of complexity into the simple experiments. And also the scale of the AI infrastructure is increasing very fast. 10,000 times per recent. For 4 years, the speed of the AI scale has been very good. And also there are many new problems. Artifacts are always moving towards GPU, storage, and CPU RAM. And nowadays PCI testing and others. We are solving the water‑like problem through the AI equipment.

[29:56] So we have been using AI OS for 10 years. [30:02] So these are the backend AI. [30:04] So some of you are already familiar with this kind of interface because we are providing our backend AI to a cloud service provider. [30:09] They are using our software to run their AI services. [30:15] So if you use some GPU cloud, you are already into a very

[30:30] Yeah, use tools, our interface. [30:33] But today, we will not use the Pegasus AI step. [30:39] But anyway, I will explain about the Pegasus AI. [30:42] We just scale the CPU with your own way and extract your data and storage with some optimizing machine technique. [30:49] And also we leverage the 13th AI iterators to provide 14 different kinds of AI iterators and deploy computing or deploy environments from the Python and TensorFlow to the VL and SushiGain and many, many kind of environments including JX and X and Bojo nowadays. [31:14] So, yeah, you can use any kind of hardware configuration on our platform. [31:23] And it's also so you can install it on your cluster or your workstation. [31:28] I can't do that. [31:29] I can't do that. [31:30] I can't do that. [31:31] I can't do that. [31:32] I can't do that. [31:33] I can't do that. [31:34] I can't do that. [31:35] I can't do that. [31:36] I can't do that. [31:37] I can't do that. [31:38] It's hard to do that. [31:39] It's hard to do that. [31:41] It's hard to do that.

[31:49] 자기네 회사들이, gpu를 관리하는 회사들이 뭐 그런 거 만들려고 하는 것입니다.

[31:42] 이렇게 말하는 것입니다.

[31:49] 자기네 회사들이, gpu를 관리하는 회사들이 뭐 그런 거 만들려고 하는 것입니다.

[31:42] 이렇게 말하는 것입니다.

[32:03] in our data data issue, which is a desktop application [32:09] to connect to the Pure AI's MPU server [32:15] to receive the features from the MPUs. [32:18] And you will have a very interesting experience today. [32:26] so the container route is a router service, not service, it's a program to routing many kind of AI/API into single AI/API. [32:35] into single AI/API. [32:39] so you can just find many services into one, like this router by yourself. [32:44] and the container hub is to manage many many container routers into a single router, router, group, groups. [32:59] and today we will use AI/API, it's actually an AI platform for your PC. [33:11] so by installing the AI on your PC, you can run your AI models on your PC using your GPUs, or MPUs, or Apple Silicons with MLSer. [33:23] or you can connect to the field AI server or UpTages server to use the cloud-based AI intelligence to power your Azure AI workflow.

[39:54] Let's have a lot of chocolate bags.

Browse all public notes