What happened
Hanmi Pharmaceutical has entered a strategic partnership with FuriosaAI, Xperix, and Bagle Labs to transition its 'Epiday' digital health platform from overseas cloud-based AI to a domestic, self-managed infrastructure. The initiative aims to localize the platform's large language model (LLM) and hardware, specifically targeting the chatbot's answer-generation function. According to the report by Chosunbiz, the companies have already completed initial testing in a validation environment to compare the accuracy and response speed of the domestic system against the existing overseas cloud setup. The transition will be implemented in phases, starting with the chatbot and eventually extending to other features like drug data searches and meal photo recognition.
Hanmi Pharmaceutical signed a strategic agreement on October 30 with FuriosaAI, Xperix, and Bagle Labs to localize the AI infrastructure supporting its Epiday digital health platform.
The project involves replacing the current overseas cloud-based AI environment with domestic servers and a Korea-style LLM to process patient health data.
Bagle Labs is responsible for app development and model replacement, Xperix will manage the domestic server environment, and FuriosaAI will provide the AI semiconductors and technical support for model optimization.
Initial testing has been conducted using a standard problem set and automated scoring framework to ensure the domestic system meets performance requirements for accuracy and speed.
The companies plan to expand the use of domestic AI technology across the platform, starting with the chatbot and moving toward more complex features like image recognition and data search.
Source details: biz.chosun.com
Why it matters
This partnership represents a significant move toward 'sovereign AI' in the South Korean healthcare sector, prioritizing data security and operational control by moving sensitive patient health information off foreign cloud servers. By integrating FuriosaAI’s domestic semiconductor technology, Hanmi Pharmaceutical aims to reduce reliance on external infrastructure, potentially lowering long-term service costs while increasing stability. The shift allows the company to maintain stricter oversight of the data used to train and refine its personalized health coaching models. This development is critical as it demonstrates a practical, industry-led application of domestic AI hardware and LLMs in a regulated, high-stakes environment like digital health, where data privacy and localized model performance are paramount. The success of this transition could serve as a blueprint for other Korean firms seeking to localize AI operations to comply with evolving data sovereignty standards.
The move addresses growing concerns regarding data sovereignty by keeping sensitive patient health information within a domestic, self-managed infrastructure.
By utilizing FuriosaAI's domestic chips, Hanmi aims to optimize costs and improve the stability of its AI services, which are currently subject to the variability of overseas cloud providers.
The partnership highlights a trend of Korean enterprises seeking to build end-to-end domestic AI stacks, from hardware to model , to gain greater control over their digital products.
The project serves as a practical test case for the efficacy of Korean-made AI semiconductors in real-world, high-demand healthcare applications.
Successful implementation could lead to more robust, personalized health coaching features, as the company gains better control over the data used to train its models.
Interactive Mechanism: How It Actually Works
Explore the underlying technology behind this development interactively.
Which component of an AI application is the machine-learning model itself?
What to watch next
The primary focus is the timeline for the full-scale deployment of the localized system into the live Epiday application. While the companies have completed validation testing, the final schedule for the transition remains unannounced. Observers should monitor the results of the 'regular working-level council' established by the four partners, as this will dictate the pace at which additional AI functions—such as meal photo recognition and drug data search—are migrated to the domestic infrastructure. Additionally, the performance metrics of the Korea-style LLM compared to the previous overseas cloud model will be a key indicator of the viability of domestic AI hardware in specialized healthcare applications.
The official date for the full integration of the domestic AI system into the live Epiday app remains unknown.
The performance of the 'Korea-style LLM' in real-world scenarios compared to the previous overseas cloud-based model will be a critical for the project's success.
The scope and speed of the planned expansion into other AI functions, such as meal photo recognition, will be determined by the newly formed working-level council.
Future updates from the council regarding quality and safety verification results will be essential to understanding the project's long-term viability.