Validated feasibility of applying domestic NPUs in real-world education settings through the “AI Semiconductor Application Demonstration Support Project”
![[엘리스그룹 사진자료] 엘리스그룹 CI.jpg](/_next/image?url=https%3A%2F%2Fcdn-cms.elice.io%2Felice-strapi%2FCI_5fd59e6ba8.jpg%3Fsp%3Dracwd%26st%3D2022-07-11T09%253A43%253A29Z%26se%3D2032-07-11T17%253A43%253A29Z%26spr%3Dhttps%26sv%3D2021-06-08%26sr%3Dc%26sig%3DXE8S5Wrx%252F6RCT1tL2PGNsACU1NvYYJAo7lDuH65zn1A%253D%26width%3D2000%26height%3D593&w=3840&q=75)
Elice Inc. (CEO Jae Won Kim), an AI full-stack company providing AI infrastructure, cloud, and industry-specific solutions, has successfully completed the 2025 “AI Semiconductor Application Demonstration Support Project.”
In this project, Elice Inc. applied an education-specialized generative AI service to domestically developed AI semiconductors whose performance the company had directly optimized to a practical, real-world level, and verified the results at scale. The demonstration is significant in that it both proves the effectiveness of domestic NPU–based AI applications in the education sector and enables the accumulation of optimization and operational know-how for domestic NPUs. This project was conducted as part of the “AI Semiconductor Application Demonstration Support Project” supported by the Ministry of Science and ICT and the National IT Industry Promotion Agency (NIPA).
Based on a prototype domestic NPU, Elice Inc. built an NPU farm and cloud infrastructure, and used it to implement AI agents for textbook information verification and education-focused Q&A. According to performance verification by an accredited testing institution, the AI application service exceeded its targets across all items in the AI application model evaluation.
The AI agent used in the demonstration is characterized by enhanced accuracy, reliability, and safety, tailored to the specific needs of educational environments. Elice Inc. constructed a high-quality Korean language–specialized dataset based on large-scale data, and on this foundation implemented:
- A Fact Checker that verifies whether content is consistent with textbook-based facts
- A Relevance Checker that blocks questions that are unrelated to textbook context or malicious in nature
- A Tone Changer that converts answers into a learner-friendly tone while maintaining grammatical accuracy
A single company building the full lifecycle from NPU farm and cloud infrastructure to AI models and services
The project stands out in that a single company directly handled the entire end-to-end process—from building AI semiconductor infrastructure and training AI models to implementing education-specialized application services. Leveraging its own AI data center, “Elice AI PMDC,” Elice Inc. independently configured servers based on domestic AI semiconductors and a dedicated cloud environment, and on top of this implemented fully integrated AI model training and service operations. Through this process, the company accumulated practical experience in performance optimization, resource allocation, and service stabilization for systems based on domestic NPUs.
“Through this demonstration, we have confirmed that education-specific generative AI agents can achieve sufficient performance and reliability even in environments built on domestic NPUs,” said Jae Won Kim, CEO of Elice Inc. “Going beyond chip adoption, we have strengthened our capabilities to implement the entire lifecycle—from AI semiconductor infrastructure to AI application services—using our own technology, based on the experience of direct optimization and large-scale demonstrations. We will continue to contribute to the expansion of the domestic AI semiconductor ecosystem and its industrial adoption.”
Looking ahead to the second year of the project, Elice Inc. plans to further improve AI agent response speed and conduct demonstrations with large-scale LLMs. Starting with education, the company also intends to expand the application of domestic NPU–based AI services across the public sector and a broad range of industries.