Category
- 🎤 Event Review
Date of issue
Apr 15, 2025
Hello 🙋🏻 This is The Invention Lab!
On Thursday, April 10th, the 2025 Startup-BuS Joint IR Program was successfully held at Pangyo 2nd Techno Valley . AI, which has grown faster than any other technology in recent years, has now established itself as a key keyword encompassing all global industries. This program, titled ' Deep Tech AI Conference ', focused on sharing the latest technology trends and insights in the global AI and deep tech industries, as well as IR presentations by excellent startups recommended by the National Creative Economy Innovation Center.
🤔 What is the Startup-BuS program?
Launched in 2024, the Center for Creative Economy and Innovation is a program of the Ministry of SMEs and Startups to support promising regional deep-tech startups, from initial investment to follow-up investment. Unlike other organizations that invest early in the year, the Center for Creative Economy and Innovation differentiates itself by conducting investments year-round.
Launched in 2024, the Center for Creative Economy and Innovation is a program of the Ministry of SMEs and Startups to support promising regional deep-tech startups, from initial investment to follow-up investment. Unlike other organizations that invest early in the year, the Center for Creative Economy and Innovation differentiates itself by conducting investments year-round.
It was a particularly informative time to broadly view the perspectives of both startups providing AI-based services and global tech companies (NVIDIA, Microsoft, Oracle) building data and AI infrastructure . In this post, I would like to summarize the key points of this event and share insights with you. 💡
1.
AI Industry Investment Status and Trends
In 2025, the overall investment market is somewhat subdued. Nevertheless, the AI sector is attracting significant market attention with exceptionally large deals, demonstrating that AI remains a powerful growth engine even amidst macroeconomic uncertainty. A prime example is the $32 billion acquisition of cybersecurity startup Wiz by Google, the largest acquisition in Google's history.
출처 : Google
Domestic AI semiconductor startup Furiosa AI was also reportedly considering a merger and acquisition (M&A) deal worth approximately 1 trillion won (approximately $850 million USD) with US tech giant Meta, but the negotiations have recently collapsed. Furiosa AI is reportedly pursuing its own path of growth and demonstrating a strong commitment to establishing itself as an alternative to NVIDIA in the global AI semiconductor market. Despite this overall weakened investment sentiment, AI and AI semiconductors remain a strategic area of interest for global tech companies and investors.
Meanwhile, interest in and use of AI is rapidly expanding not only in the industrial sector but also among general consumers. Did you know that Korea is one of the top three countries with paid ChatGPT users? Furthermore, the number of domestic users utilizing the Ghibli-style image creation feature recently surpassed 5 million. This demonstrates that AI is moving beyond mere technology and becoming a part of everyday experiences.
지브리 스타일로 만든 강연 사진 | 출처 : 더인벤션랩
2.
Global Expansion Lecture
2-1. [Oracle] Oracle's Perspective on Artificial Intelligence Technology
Key words: AI Agent, SW Defined X(SDDC, SDV, SDM), Physical AI, Digital Twin, Data
•
As technological advancements in Large Language Models (LLMs) accelerate, the spectrum of models available to users is expanding. However, going forward, the ability to strategically select the "right model" for a given situation will likely become more important than simply selecting a more accurate or up-to-date model.
◦
In ultra-precision fields where absolute errors are unacceptable, rule-based systems may be more suitable than generative AI.
◦
Depending on what data you will be using, you will be able to organize your requirements and map out which technologies you will use later.
출처 : 당근 테크 블로그
•
In terms of AI infrastructure, the importance of not only hardware but also the software that controls and optimizes it is growing. Oracle, for example, is striving to move beyond its traditional database identity and become a cloud provider. Efficient data center management and structural visibility are emerging as crucial challenges. Many companies currently lack visibility into how their data centers are managed and structured, and in this context, the need for SDM is emerging.
•
As AI advances, the importance of integrated management capabilities for structuring, updating, and operating underlying data is expected to grow.
2-2. [Microsoft] AI and AI Trends as Perceived by Microsoft
Key Keywords: Security, Responsible AI, Easy to Build, Integration
•
While Microsoft views AI as a key strategic keyword, it is more fundamentally interested in the underlying technologies that form the foundation of technology.
•
These days, when a single technology emerges, dozens of similar technologies emerge within days. Speed and execution are becoming far more important factors than the technology itself. In this context, the value of "Easy to Build," which allows anyone to quickly implement a solution, is continually being highlighted.
AI Azure Studio | 출처 : AI 타임스
•
According to a survey conducted by Microsoft, 95% of companies are already utilizing AI or considering introducing it, but less than 1% of them have established a “safe AI environment.”
•
As LLM-based technologies proliferate, numerous endpoints and complex technical environments are simultaneously created, making security design and control more challenging than ever.
◦
Especially for startups aiming for global expansion or collaboration with large corporations, it's essential to incorporate a systematic approach to security from the initial prototype stage. Indeed, despite startups boasting superior technology and business potential, security issues frequently lead to breakdowns in collaboration or investment negotiations.
3.
Introducing AI and Big Data Startups
•
Established: November 2017
•
Industry: Bio/Medical, AI/Deep Tech/Blockchain
•
Business Item: AI-based sign language translation platform
•
Company CEO: Kim Jong-hwa
•
Headquarters location: 706 Eonju-ro, Gangnam-gu, Seoul
•
Main business
◦
AI-based sign language translation service: HandsignTalk
→ Text or voice is translated into sign language and expressed in real time by an avatar, improving information accessibility for the hearing impaired.
→ B2G, B2B kiosk and mobile application operation
→ We have launched a sign language book in collaboration with Kyobo Book Centre and are also pursuing collaboration with the insurance/finance industry.
→ Text or voice is translated into sign language and expressed in real time by an avatar, improving information accessibility for the hearing impaired.
→ B2G, B2B kiosk and mobile application operation
→ We have launched a sign language book in collaboration with Kyobo Book Centre and are also pursuing collaboration with the insurance/finance industry.
◦
Korean-Korean Sign Language Parallel Corpus Construction Project
→ Building core data for video-based sign language recognition
→ Building core data for video-based sign language recognition
◦
AICC platform
→ Collect various data (text, voice, images) and convert them into data that AI can learn from.
→ AICC integrated platform service configured to interconnect and integrate independent AI engines for efficient AI consulting.
→ Collect various data (text, voice, images) and convert them into data that AI can learn from.
→ AICC integrated platform service configured to interconnect and integrate independent AI engines for efficient AI consulting.
•
Business Model: B2G/B2B deployment, platform usage fee (subscription), custom solution development fee
•
Investment round: Series A
•
Company Name: Postmass
•
Established: April 2021
•
Industry: Education/Mathematics Education
•
Business Item: Math Content Creation Platform
•
Company CEO: Choi Jun-ho
•
Headquarters Location: 28 Digital-ro 30-gil, Guro-gu, Seoul
•
Main business
◦
Math Content Creation Platform: Math Secretary
→Currently, it is a platform that allows math teachers to easily create and edit worksheets for instructors.
→ By registering school exam papers, answers, and internal exam answer sheets in a database, you can generate revenue by sharing a certain portion of the sales whenever the database is sold.
→ All technologies within the platform (OCR, graph generator, etc.) are developed in-house.
→ It is steadily generating user growth and sales without any special advertising.
→Currently, it is a platform that allows math teachers to easily create and edit worksheets for instructors.
→ By registering school exam papers, answers, and internal exam answer sheets in a database, you can generate revenue by sharing a certain portion of the sales whenever the database is sold.
→ All technologies within the platform (OCR, graph generator, etc.) are developed in-house.
→ It is steadily generating user growth and sales without any special advertising.
•
Business Model: Freemium, Content Revenue Sharing, Subscription
•
Investment round: Series A
•
Company Name: Korea Education Partners
•
Established: April 2024
•
Industry: Education, AI/Deep Tech/Blockchain
•
Business Item: AI-based Life Record Analysis Service
•
Company Representatives: Kwon Ki-won, Kwon Jae-woo
•
Headquarters Location: 21 Beonyeong-ro 124beongil, Nam-gu, Ulsan
•
Main business
◦
Student records and detailed ability and special skills management solution: Hakjong Pro
→ AI-based SaaS platform
→ Automatically generates special content tailored to the student's career path and aptitude.
→ AI trained on the evaluation criteria of actual admissions officers to analyze students' academic records and help them set goals.
→ Used by over 1,300 academies nationwide
→ AI-based SaaS platform
→ Automatically generates special content tailored to the student's career path and aptitude.
→ AI trained on the evaluation criteria of actual admissions officers to analyze students' academic records and help them set goals.
→ Used by over 1,300 academies nationwide
◦
B2C Customized Extracurricular Activity Management Solution: Seteuk Yeo-Wang
→ Recommend extracurricular activities tailored to students' interests and career paths
→ Activity history management and automatic portfolio creation
→ Recommend extracurricular activities tailored to students' interests and career paths
→ Activity history management and automatic portfolio creation
◦
Data Collection/Processing Platform for AI Learning: LinAI
→ Collection and processing of AI learning datasets related to education/knowledge based on a carefully selected pool of experts.
→ Build an efficient data processing system through a labeling automation engine.
→ Collecting and processing AI learning datasets required by 26 education/knowledge service companies.
→ Collection and processing of AI learning datasets related to education/knowledge based on a carefully selected pool of experts.
→ Build an efficient data processing system through a labeling automation engine.
→ Collecting and processing AI learning datasets required by 26 education/knowledge service companies.
•
Business Model: SaaS subscription, data-driven solution licensing
•
Investment Round: Seed
•
Company Name: Quasar
•
Established: June 2020
•
Industry: Fashion, AI/Deep Tech/Blockchain
•
Business Item: AI-based luxury goods counterfeit detection solution and luxury goods commerce platform.
•
Company CEO: Kim Myeong-hyeon
•
Headquarters Location: 293 Bupyeong-daero, Bupyeong-gu, Incheon
•
Main business
◦
AI-based authenticity/counterfeit identification service: Quaiza
→ Collect and standardize over 300,000 genuine and fake data (materials, logos, stitching, metal reflections, etc.) to train AI models.
→ Discrimination of genuine/fake products through a complex data inference model based on a mixed group of experts' MOE model
→ It takes about 3 to 5 minutes and has 95% accuracy (human feedback is in progress for the remaining 5%)
→ Collect and standardize over 300,000 genuine and fake data (materials, logos, stitching, metal reflections, etc.) to train AI models.
→ Discrimination of genuine/fake products through a complex data inference model based on a mixed group of experts' MOE model
→ It takes about 3 to 5 minutes and has 95% accuracy (human feedback is in progress for the remaining 5%)
◦
B2G service to be provided
→ A new model capable of detecting drugs by smell is planned to be developed.
→ A new model capable of detecting drugs by smell is planned to be developed.
•
Business Model: Subscription-based, per-transaction fee
•
Investment Round: Seed
•
Company Name: GSF Solution
•
Established: May 2022
•
Industries: Manufacturing/Hardware, AI/Deep Tech/Blockchain
•
Business Item: FDC solution specialized for each semiconductor manufacturing process
•
Company Representative: Seunghwan Kim
•
Headquarters location: 97 Gakri 1-gil, Ochang-eup, Cheongwon-gu, Cheongju-si, Chungcheongbuk-do
•
Main business
◦
FDC Solutions
→ Collect/classify/analyze data on static electricity, temperature and humidity, flow rate, etc. generated in the semiconductor manufacturing process to monitor equipment abnormalities and diagnose accidents to prevent them.
→ Connect up to 40 sensors to a single module to integrate and manage various sensor information into a single module.
→ Customized solutions for each semiconductor manufacturing process and accident
→ Collect/classify/analyze data on static electricity, temperature and humidity, flow rate, etc. generated in the semiconductor manufacturing process to monitor equipment abnormalities and diagnose accidents to prevent them.
→ Connect up to 40 sensors to a single module to integrate and manage various sensor information into a single module.
→ Customized solutions for each semiconductor manufacturing process and accident
◦
On-device data analytics platform
→ Maintain optimal equipment performance through algorithm-based data processing.
→ Maintain optimal equipment performance through algorithm-based data processing.
•
Business Model: Initial Build, Subscription, Technology Maintenance Contract
•
Investment Round: Seed
•
Company Name: Care Six
•
Established: May 2019
•
Industry sectors: Home living/pets, AI/deep tech/blockchain
•
Business Item: IoT-based pet wearable
•
Company CEO: Kim Seong-hun
•
Headquarters Location: 2870, 516-ro, Jeju-si, Jeju
•
Main business
◦
IoT-based dog wearable
→ Type 1: Vet > Vital Sign Monitoring System for Animal Hospitals
→ Type 2: Guardian > A monitoring system for guardians to use at home after discharge.
→ Type 3: Home > Pet Health Monitoring Devices for General Consumers
→ The world's first technology that combines ballistic cardiography (BCG) and a 6-axis sensor (enabling early prediction of most diseases, except those requiring blood tests)
→ Measures biosignals with over 90% accuracy without hair removal or electrode attachment
→ Algorithm based on over 300,000 clinical data
→ Type 1: Vet > Vital Sign Monitoring System for Animal Hospitals
→ Type 2: Guardian > A monitoring system for guardians to use at home after discharge.
→ Type 3: Home > Pet Health Monitoring Devices for General Consumers
→ The world's first technology that combines ballistic cardiography (BCG) and a 6-axis sensor (enabling early prediction of most diseases, except those requiring blood tests)
→ Measures biosignals with over 90% accuracy without hair removal or electrode attachment
→ Algorithm based on over 300,000 clinical data
•
Business Model: Build-to-Order, Subscription, Data-Driven Platform Monetization
•
Investment Round: Seed
•
Company Name: Muse Blossom
•
Established: May 2021
•
Industry: Content/Arts
•
Business Item: Audio Watermarking Technology-Based Content Source Verification and Copyright Protection Service Platform
•
Company Representative: Eunseon Jo
•
Headquarters location: 84 Madeul-ro 13-gil, Dobong-gu, Seoul
•
Main business
◦
Audio watermarking
→ Protect copyright by inserting a specific message into the audio (using transient-based sidechain audio watermark coding technology that hides the watermark in the audio source frequency peaks)
→ Protect copyright by inserting a specific message into the audio (using transient-based sidechain audio watermark coding technology that hides the watermark in the audio source frequency peaks)
◦
Deep Voice Detection Solution
→ Deep Voice prevents damage by sending alerts to phone users during calls even when one-way phishing is detected.
→ Detection rate of 90% or higher
→ Deep Voice prevents damage by sending alerts to phone users during calls even when one-way phishing is detected.
→ Detection rate of 90% or higher
•
Business Model: Build-to-Order, Subscription, License Sales
•
Investment Round: X
•
Company name: Conexi
•
Established: February 2022
•
Industry: AI/Deep Tech/Blockchain
•
Business Item: Decentralized Secure File System Storage Platform
•
Company Representative: Choi Young-il
•
Headquarters Location: 524 Bongeunsa-ro, Gangnam-gu, Seoul
•
Main business
◦
Decentralized secure file system storage platform
→ Provides data protection solutions such as system status monitoring and control, real-time notifications, real-time backup and recovery, and ransomware response through proprietary technologies such as computing servers, storage (cloud), and content hash access.
→ 0% chance of data leakage > Even if the content is split into pieces and stolen by content identifier, the contents cannot be viewed.
→ Provides data protection solutions such as system status monitoring and control, real-time notifications, real-time backup and recovery, and ransomware response through proprietary technologies such as computing servers, storage (cloud), and content hash access.
→ 0% chance of data leakage > Even if the content is split into pieces and stolen by content identifier, the contents cannot be viewed.
•
Business Model: Build-to-Order, Subscription
•
Investment Round: Seed
4.
Talk Concert Summary
1️⃣ Domain-Specific Strategy : When developing and deploying AI models, considering domain characteristics is paramount. Especially in industries where factual accuracy is paramount (e.g., healthcare, law, etc.), fine-tuning is necessary to intentionally suppress creativity and encourage fact-based responses to reduce hallucinations in generative AI. Furthermore, it may be more appropriate to exclude generative AI altogether and instead apply search-based AI.
2️⃣ The importance of rapid market entry : As AI rapidly advances, speed is crucial for startup success. This requires easy-to-build functionality, and flexible integration across diverse environments can be a competitive advantage.
3️⃣ Operations are more important than design : As data grows and the number of concurrent users increases, various operational bottlenecks arise, such as model inference speed, API response latency, GPU utilization, and cluster load. Therefore, it's important to consider how to efficiently control them from an operational perspective.
4️⃣ When expanding globally, consider regulatory diversity : AI ethics and data-related regulations vary widely across countries and regions. This issue is particularly acute in industries like healthcare, which handle sensitive information. For example, in the US, health care laws vary by state. While some states allow users to freely utilize their data, others restrict even users' access to their own medical data. Therefore, establishing a legal compliance system is a key component.
5️⃣ The direction of AI technology : With the emergence of DeepSearch, a trend has emerged toward creating cost-effective AI solutions through AI lightweighting and optimization. Nevertheless, big tech companies are still moving toward building large-scale, high-performance infrastructure, and this direction can be chosen based on a company's specific circumstances.
5.
Final Insights
This conference taught us that while generative AI is a global trend, it's not universally applicable to all industries. Specifically, because the " importance of facts " varies across industries , it's essential to first assess the importance of facts in a given industry before implementing AI.
Furthermore, while past AI research focused on technology primarily in terms of "accuracy" and "applicability," looking at it from the perspective of planning actual services and providing infrastructure, I realized that " operability " and "system efficiency " are emerging as far more important factors. Going forward, designing a balance between model performance and operational efficiency will likely become a key competitive advantage.
Finally, I was particularly impressed by the differences in regulations across countries and regions. Beyond simply developing technology, I realized that operating it as a real service requires comprehensive consideration of not only the scope of available data, model performance, and system efficiency, but also the legal and ethical regulations of the respective region.
Beyond AI technology, the ability to comprehensively design operational environments and institutional factors is becoming a key competitive advantage. Going forward, strategic decisions about where to apply and how to operate efficiently will likely make a greater difference than the technology itself.
현장 사진 | 출처 : 더인벤션랩
Subscribe to 'The Invention Lab Blog'
If you subscribe to the site, you can be the first to receive the latest updates, including new posts, via notifications and email.
Join Slashpage and subscribe to 'The Invention Lab Blog'!
Join Slashpage and subscribe to 'The Invention Lab Blog'!
