[Dongguk University Campus Town x Scale-Up] Dongguk University Campus Town is conducting the '2025 Scale-Up Program' in collaboration with the Scale-Up team. Dongguk University Campus Town and the Scale-Up team will introduce the overall business operations of the startups participating in this program and analyze their business models. They aim to assist these startups in solving challenges for the next stage by connecting them with experts from various fields.

AnsimHi is an AgeTech company that improves welfare services using technologies such as Speech-to-Text (STT) and Artificial Intelligence (AI). AgeTech is a compound word of Age and Technology, referring to products, services, and technologies that enhance the quality of life for the elderly.

Currently, AnsimHi is developing a solution that records and automatically summarizes conversations between care workers, welfare counselors, and welfare recipients using voice recognition and STT technology tailored for the elderly. This aims to reduce the workload of care workers and proactively provide necessary services to welfare recipients.

To enhance the quality and effectiveness of its services, AnsimHi utilizes various technologies such as elderly-specialized STT, AI lightweighting, and security. AnsimHi explains that they are advancing their services by applying new technologies to strengthen their position as an AgeTech company.

Met with Kang Byung-min, the developer in charge of AI development at AnsimHi, to discuss the technological capabilities secured by AnsimHi and future advancement plans.


Attracted to AnsimHi for solving care field problems with technology

IT Donga: Hello, Mr. Kang Byung-min. Please introduce yourself.

Kang Byung-min: Hello, I'm Kang Byung-min from AnsimHi. I am currently in my fourth year, second semester, at Dongguk University, majoring in Multimedia Software Engineering, and I am in charge of AI development at AnsimHi.

I have participated in the International Collegiate Programming Contest (ICPC) as a school representative and have consistently participated in AI hackathons. I received the Excellent Paper Award at the Summer 2025 Korean Institute of Information Technology Conference for my research in phonetics and won the Excellence Award at the Information and Communication Society Idea Contest with a language translation model using STT technology.

IT Donga: When did you join AnsimHi, and what was the reason for joining?

Kang Byung-min: I joined AnsimHi during its startup preparation phase, having worked on a project together with CEO Kim Min-soo during university. I officially joined after the startup.

I have always been interested in improving inefficient processes, and I was impressed by how AnsimHi solves problems in the care field using IT technology. I was particularly attracted to the idea of addressing issues in the growing senior market, which led me to join.

IT Donga: What kind of company do you think AnsimHi is?

Kang Byung-min: AnsimHi is a company that creates services you can 'rest assured' with. It aims to create an environment where both people and society can feel secure by solving real field problems, not just focusing on technology. Personally, it is a place where I can grow as a developer, achieve results with good people, and create value. I find it enjoyable to work with local governments to impactfully improve senior market issues and experience these improvements firsthand.

Differentiation with elderly-specialized STT, AI, and de-identification technology

IT Donga: Why does AnsimHi emphasize technological capabilities?

Kang Byung-min: The elderly population is rapidly increasing, but the workforce to care for and manage them is decreasing. I believe AgeTech technology is the most realistic way to solve this problem.


IT Donga: Please introduce AnsimHi's technological capabilities.

Kang Byung-min: As an AgeTech company, AnsimHi applies technologies such as elderly-specialized STT and AI to contribute to the digital transformation in the welfare sector. We are developing three main technologies.

The first is STT technology. It is optimized for elderly speech by building a dataset reflecting elderly speech characteristics and algorithms reflecting elderly conversation characteristics. Our STT model has a higher sentence generation accuracy based on elderly conversations than open-source-based general STT models.

Another feature is that it operates without external APIs. Since our service handles personal and sensitive information of the elderly and care workers, security is crucial. Therefore, we use AI models that operate in an environment blocking external leaks without using big tech companies' large language models (LLMs). Thanks to this, local governments and public institutions can use it with peace of mind without worrying about data leaks.

The second technology is a care-specialized LLM. It recommends customized policies by connecting to the welfare service database (DB) of the government and local governments based on Retrieval-Augmented Generation (RAG). Developed based on open-source models, it can be applied in counseling fields with strict security.

The third is de-identification technology. It automatically detects and processes personal and sensitive information in compliance with personal information protection regulations. We have built a pipeline that automatically masks information that can identify individuals, such as resident registration numbers, addresses, and phone numbers.

IT Donga: How do these technologies affect AnsimHi's services?

Kang Byung-min: They improve the practical work of care workers. Conversations held on-site are recorded with high accuracy and summarized into reports, reducing the burden of report writing. Welfare recipients can receive the services they need more quickly, alleviating inconveniences in their lives. Decision-makers in local governments or public institutions can increase productivity and performance by adopting AI without concerns about personal information leaks.

Service improvement through technological advancement, developing lightweight models

IT Donga: Please tell us about the direction for adding new technologies or advancing current technologies.

Kang Byung-min: We are advancing our services by applying new technologies. We are focusing on lightweighting AI models. Currently, we provide services via the cloud, but we are developing lightweight models that enable AI processing on care workers' smartphones.


We plan to achieve lightweighting through internal model separation and domain-specific knowledge assignment. AI services go through numerous computational processes, and we aim to lightweight the entire system by separating each process and applying AI only where necessary. We also plan to introduce sLLM (small Large Language Model) lightweighted to the extent that it can run on a general PC.

Additionally, we aim to achieve lightweighting by further advancing dictionary algorithms using elderly conversations, dialects, and welfare-related terms, specializing in the social security field. We plan to continuously advance by accumulating more data through upcoming demonstration projects.

Furthermore, we plan to add a real-time speaker separation feature to STT technology. We are implementing real-time speaker separation by pre-receiving the voices of care workers or welfare counselors before conversations to secure standard voice data. This method reduces computation and improves accuracy. We have completed it for internal testing and are currently correcting errors and advancing it. We plan to complete the final version by October and introduce it in demonstration projects to be conducted in the future.

To enhance data security, we also plan to advance de-identification technology. We are developing a full-range data encryption and de-identification framework where even data managers cannot view the original data. This method de-identifies data on user devices, storing only information without personal data on the server. This aims to fundamentally block personal information leaks.

For reference, we have implemented a framework combining the lightweighting and de-identification modules developed so far and submitted it to the Ministry of Science and ICT's 2025 GovTech Startup Contest. We have advanced to the finals, and the final results are scheduled to be announced in November.

IT Donga: Lastly, please tell us about your future plans.

Kang Byung-min: We are about to conduct demonstration projects with local governments, and we aim to complete them successfully and achieve good results. I want to contribute to AnsimHi's rapid growth by continuously researching technologies that align with AnsimHi's long-term direction.

IT Donga Han Man-hyuk (mh@itdonga.com)

Popular News