KARI to Develop 'AI Brain'... 300 Billion-Parameter Model Also Operated In-House

Utilizing AI Without Sending Research Data Outside
KARI BRAIN Platform Set for Launch

Korea Aerospace Research Institute (KARI) is developing its own "AI brain" that will allow artificial intelligence (AI) to search and utilize decades of accumulated aerospace research data and researchers' experiences. KARI plans to internally operate even large-scale AI models with over 300 billion parameters, advancing AI transformation (AX) across research and development, as well as administrative operations.

On September 14, KARI announced that it will build an intelligent platform called "KARI BRAIN" to connect its accumulated aerospace R&D knowledge and data with AI.

Main building exterior view of the Korea Aerospace Research Institute. Provided by the Korea Aerospace Research Institute

The core of KARI BRAIN is to convert technical documents and research experiences—currently scattered across individuals and departments—into knowledge assets that AI can utilize. KARI aims to gradually link verified materials to develop the platform into a tool that supports finding necessary information for R&D and solving problems.

KARI has also established its own AI infrastructure. In January this year, it introduced a server equipped with four NVIDIA H200 graphics processing units (GPUs), and in April, it added another server with eight B200 GPUs. The institute now operates several AI models internally, including a large-scale model with over 300 billion parameters. This setup enables researchers to utilize AI without sending sensitive research data outside the organization. Currently, there are 836 users of the in-house generative AI services.

In the R&D sector, AI usage will expand to include the analysis of technical data, support for design and analysis, processing of test data, and software development. For administrative operations, KARI plans to leverage AI for regulation searches, document drafting, and automation of repetitive tasks.

To this end, KARI held its first meeting on September 14 for 13 "AX Leaders" selected from eight departments. Researchers and operational teams are directly identifying AI utilization tasks they need, while the AX Promotion Office—established in July and reporting directly to the vice president—will support the distribution of shared technologies and tools.

KARI has designated the period until July next year as the first AX implementation phase. After measuring the effectiveness of field application, the institute plans to expand proven AI usage methods and tools organization-wide. A management system will also be established to protect research data and ensure the reliability of AI outputs.

Lee Sangcheol, President of KARI, stated, "We will connect KARI's accumulated expertise and research data with AI to reinforce the professionalism of our researchers." He added, "By reducing the burden of repetitive tasks and creating an environment that allows greater focus on core research, we aim to enhance the competitiveness of aerospace research and development."

This content was produced with the assistance of AI translation services.