TeamSparta Ranks in Top 20 of NVIDIA AI Challenge, Unveils Air-Gapped “AX Fortress”
TECHWORLD ·
✦ AI Summary
TeamSparta was selected as a Top 20 finalist for the NVIDIA Inception Grand Challenge 2026 while having its own in-house closed-network AI solution, AX Fortress.
The finals were held in Singapore on the 22nd, and TeamSparta unveiled the AX Fortress-based “Supply Chain Volatility Engine.”
The engine does not send corporate data to external cloud services in an Air-Gapped environment, detects parts supply risks 30 to 60 days in advance, and has AI agents that search and analyze the right ordering time, alternative suppliers and compatible parts.
TeamSparta, which has its own in-house air-gapped AI solution, “AX Fortress,” has advanced to the finals of an NVIDIA APAC AI startup competition. TeamSparta was selected as a Top 20 finalist for the “NVIDIA Inception Grand Challenge 2026,” with the announcement made on the 23rd.
The challenge was conducted in a way that had AI startups in the Asia-Pacific region solve industry problems presented by global companies using AI technology. About 500 companies from Korea, Singapore, Taiwan, Australia and other markets participated, and TeamSparta advanced to the Top 20 through the Korea semifinals.
The finals were held in Singapore on the 22nd. At the finals, TeamSparta unveiled the AX Fortress-based “Supply Chain Volatility Engine.”
The “Supply Chain Volatility Engine” is based on an Air-Gapped environment that does not send corporate data to external cloud services. The engine detects parts supply risks 30 to 60 days in advance, and it is equipped with functions that allow AI agents to search and analyze the right ordering time, alternative suppliers and compatible parts.
AX Fortress was designed as an internal operation structure that does not allow sensitive corporate data to be taken outside, and it is configured to run AI models, search functions and agents simultaneously. The system includes its own LLM and integrates knowledge-graph-based RAG and AI agents, with the operating environment kept inside a closed network. It is intended for tasks such as supply chain work that require identifying relationships across multiple data sets.
AX Fortress uses NVIDIA DGX Spark as its hardware foundation and applies NVIDIA NVFP4 4-bit quantization to its in-house model. As a result, the company built an integrated system that does not require a separate large server room or air-conditioning equipment and can operate in an office environment.
The system supports a 1 million token context. The company said its inference speed exceeds 50 tokens per second (50+ TPS). That is more than 5 times faster in processing performance than roughly 10 tokens per second seen in comparable air-gapped open-source models.
TeamSparta said it improved its AI agent functions to boost performance in file handling, search and the use of various tools. In TeamSparta's own testing, the Tool Call success rate was 100%.
TeamSparta also applied security features designed for closed-network environments. It implemented an Air-Gapped structure that blocks external communications, a physical seal to prevent unauthorized opening of the hardware, SSH blocking and SHA-512-based hash chain audit logs so that equipment access and work history can be verified.
TeamSparta applied its own AI model optimization technology to AX Fortress. It also expanded the model training and optimization technology it secured during the process of its in-house lightweight model taking first place in the HLE(Ko) category of the K-AI leaderboard run by the Ministry of Science and ICT and the National Information Society Agency (NIA) into an enterprise closed-network AI environment.
TeamSparta plans to use the challenge as a springboard to expand AX Fortress beyond supply chain management to security monitoring, manufacturing process analysis, R&D and coding automation. It is currently carrying out PoC and on-site deployment in network and communications security monitoring, finance, electronic component manufacturing and chemical materials research, focusing on industries that share the common trait of operating in environments where external data export is restricted.
Lee Beom-gyu, CEO of TeamSparta, said the challenge confirmed on a global stage the possibility that closed-network AI can solve real industrial problems. He added that the company plans to continue advancing AX Fortress technology to support task automation using internal data even in environments constrained by security and infrastructure limitations.
Source: TECHWORLD · Kim Seung-gi
Original: https://www.epnc.co.kr/news/articleView.html?idxno=407346
References
This article was produced with the help of an automated content generation algorithm.
Source: TECHWORLD
View originalThis article was summarized and organized by BizCrush based on the original article from TECHWORLD. For exact quotations and full details, please refer to the original article.