Korea to get medical LLM tailored to local clinical practice

A large language model tailored specifically to Korea's medical field is set to be developed.

Foreign AI models have been criticized for failing to adequately reflect the context and requirements unique to Korean medical settings. These include Korean-language medical terminology, prescription and reimbursement code systems, Health Insurance Review and Assessment Service standards, Ministry of Health and Welfare notices, and personal medical information protection rules. Concerns have also been raised about data leaks, as sensitive medical information on Korean patients could end up flowing overseas.

Medical information company BITComputer announced Wednesday that it had signed a memorandum of understanding with Amazon Web Services to build a medical-specialized LLM and migrate to the cloud.

Under the agreement, BITComputer will combine 43 years of accumulated proprietary data and clinical know-how with AWS's cloud and AI technology. The company plans to use this to build its own LLM optimized for Korea's medical environment — something general-purpose models alone struggle to adequately address.

BITComputer plans to train the model on its accumulated, de-identified electronic medical record data along with domestic medical literature and clinical guidelines. It aims to develop this into a common foundation model to be shared across its entire product lineup and future Korean medical AI agents. Over the past 43 years, the company has built up data and know-how from Korea's medical field. It has done so through its EMR systems for hospitals and clinics, digital health care services and IT education business.

"We currently operate our own AI product lineup, including BitMate, an AI-based EMR system for clinics; MediGent AI, an AI tool supporting insurance review; and DrugInfo AI, an AI tool for drug information," the company said. "We plan to combine this data and know-how with AWS's cloud and AI model capabilities to build an LLM specialized for Korea's medical field."

The two companies agreed to cooperate in four areas. They will develop a Korea-specialized medical LLM trained on de-identified domestic EMR data and medical literature. They will also upgrade the clinical and administrative support functions of existing AI products such as BitMate AI and MediGent AI. The partnership further covers expanding voice-recognition-based automated charting for clinical settings and migrating AI model development, training and operating environments to the AWS cloud.

"We will build a Korea-specialized medical LLM that renowned foreign models cannot replicate," said Cho Jae-suk, BITComputer's chief strategy officer. "By securing a cloud-based serving environment together, we will raise the quality and speed of our entire product lineup to the next level."

"As Korean medical institutions apply AI to a growing range of tasks — from writing medical records to supporting insurance review and checking drug information — demand is rising for specialized models that accurately reflect Korean clinical terminology and the country's regulatory framework," said Cho Min-sung, head of AWS Korea's health care business. "We will develop a medical-specialized LLM to create an environment where medical staff can focus more on treating patients."