Kreverse Selected as Demonstration Partner for FuriosaAI RNGD

Participation in National Strategic Project as a Leading Company Driving Educational AX Innovation

Kreverse, a leading domestic edtech company led by CEO Lee Donghoon, announced that it has been selected as a demonstration partner utilizing the NPU-based infrastructure of AI semiconductor company FuriosaAI in the 'AI Computing Demonstration Infrastructure Advancement' project, overseen by the National IT Industry Promotion Agency (NIPA). Kreverse will participate in this project together with AI service platform company RYUTEN Technologies (hereafter 'RYUTEN'), verifying the potential applications of FuriosaAI's second-generation chip, RNGD (Renegade), in different AI service environments.

The 'AI Computing Demonstration Infrastructure Advancement' project is a national strategic initiative with a total budget of 35.5 billion KRW over three years, running from 2025 to 2027. Its objective is to verify the performance and applicability of domestically produced AI semiconductors by applying them to real-world service environments. In this demonstration, Kreverse plans to apply RNGD-based AI infrastructure to educational services, focusing on verifying response speed, throughput, and power efficiency compared to the conventional GPU-centric infrastructure.

Educational services are high-load environments where a large number of user requests and real-time AI computations occur simultaneously, making the stability and scalability of AI infrastructure critically important. Kreverse plans to verify the feasibility of stable AI service operation in large-scale concurrent access environments, using actual learning data generated from its nationwide network of approximately 200 academies. The fact that the demonstration will be conducted in real educational settings, rather than a single testbed, makes the results particularly noteworthy.

This demonstration involves RYUTEN, representing generative AI application services, and Kreverse, representing educational services. Both companies will apply NPU-based AI infrastructure to their respective large-scale generative AI and educational service environments, verifying performance and stability under actual user traffic and computational load conditions. Through this, they aim to determine whether NPU-based infrastructure can expand beyond specific industries into various AI service domains.

In this demonstration project, Kreverse will focus on its AI-based English writing and speaking automated evaluation service, HUMMINGo. HUMMINGo analyzes speaking and writing results in real time, providing rubric-based scores and personalized feedback, and is already being stably operated in educational settings.

During the demonstration process, approximately 10 to 20 percent of the AI traffic currently running on Azure will be switched to the RNGD-based infrastructure, and the LG EXAONE 4.0 model will be applied to verify specific service metrics. EXAONE is a generative AI model independently developed by LG AI Research, which was recently ranked seventh in the world (open weight category) by the global AI performance evaluation agency Artificial Analysis. In July, LG AI Research announced the adoption of RNGD for EXAONE.

Kreverse distinguishes itself by not merely introducing AI technology into educational settings, but by applying it comprehensively to actual learning experiences and operational structures. Its structure, which combines AI automated evaluation, feedback based on learning data, and nationwide operational experience, is considered a rare case in the education sector. This demonstration is expected to serve as an important reference for verifying the potential expansion into public education, not limited to private education. Through the nationwide educational service environment demonstration, it will also be possible to verify effectiveness on a scale extending beyond individual schools to city and provincial education offices.

Kim Gwan, Executive Director of Kreverse's AX Division, stated, "This demonstration is significant in that it verifies FuriosaAI's NPU-based infrastructure in an actual educational service environment," adding, "Together with RYUTEN, we will contribute to proving the practical applicability of domestic AI infrastructure in various AI service domains."