TwoDigits Open-Sources K-Judge Korean Evaluation AI

Domestic AI company TwoDigits is drawing significant attention from the AI industry by unveiling 'K-Judge', its proprietary system capable of precisely evaluating the performance of Korean Large Language Models (LLMs). K-Judge is a self-developed solution designed to deeply analyze Korean responses generated by LLMs in terms of context and grammatical accuracy, thereby highly evaluating the model's completeness and natural language processing capabilities. Particularly, the most noteworthy feature of this system is its support for an offline environment that operates perfectly without an internet connection. This provides a strong basis for its high value in high-security industries where internal verification without reliance on external services is essential, such as defense, finance, and healthcare. In fact, it has been confirmed that the adoption of K-Judge is being actively reviewed in an AI support system development project currently underway by TwoDigits with Korea Hydro & Nuclear Power (KHNP), proving its practical utility. In the development process of K-Judge, the full support of the Gwangju AI Industry Convergence Agency (AICA) played a decisive role. Thanks to AICA's high-performance GPU resources, TwoDigits was able to further strengthen its Korean dataset and create synergy in advancing LLM technology. Both organizations plan to continue close strategic cooperation in the future, aiming to foster a domestic AI industry ecosystem and achieve technological self-reliance. AI industry officials unanimously agree that "the technology to accurately and reliably verify developed models" is now emerging as a core competitive advantage, as much as "the technology to develop AI models." Particularly in critical industrial sectors where sensitive information is exchanged, such as public administration, finance, and healthcare, the use of external AI solutions is often restricted due to security reasons, making the need for a domestic evaluation system that can independently measure and assess the performance of internally developed models increasingly imperative. Park Seok-jun, CEO of TwoDigits, stated, “K-Judge is an innovative technology, optimized for the characteristics of Korean data as an LLM evaluation model, that can provide practical assistance in various development environments.” He expressed a strong commitment, adding, “We will continue to strive to strengthen the competitiveness of domestic AI technology and build a self-reliant ecosystem.”