- Secured the top position in Track C, alongside the Grand Prize among 20 competing teams, utilizing synthetic data generation technology tailored for Mixture of Experts (MoE) quantization.
- Developed a dataset utilizing an agent grounded in the Nemotron 3 Super120B, embracing a data-centric approach instead of an algorithm-centric optimization methodology.
SEOUL, South Korea, April 24, 2026 /PRNewswire/ — Nota AI, a preeminent company in AI model compression and optimization, announced its triumph in Track C at the NVIDIA Nemotron Hackathon, employing advanced synthetic data generation technology specialized in MoE quantization. Notably, it was conferred the Grand Prize, affirming its excellence among all 20 competing teams.
The hackathon was meticulously orchestrated to disseminate the latest advancements surrounding NVIDIA’s groundbreaking open-source AI model, “Nemotron”, while providing Korean developers with the opportunity to gain firsthand experience with state-of-the-art AI technologies.
Contestants were divided into three tracks: developing AI agents to tackle real-world challenges (Track A), enhancing domain-specific Nemotron models via SFT and RL (Track B), and crafting synthetic data pipelines to create premium datasets (Track C).
Nota AI emerged victoriously in Track C, dedicated to synthetic data pipelines for sustainable development goals (SDG), garnering the highest overall evaluation, subsequently securing the Grand Prize.
During the competition, Nota AI deployed an agent based on NVIDIA’s Nemotron 3 Super model to construct a quantization dataset tailored for the MoE architecture.
This initiative showcased a data-centric strategy that markedly diminishes performance degradation in MoE models.
In contrast to traditional quantization techniques that predominantly emphasize formulaic and algorithmic optimization, Nota AI distinguished itself by meticulously crafting the structure, caliber, and task alignment of its dataset, leading to enhanced quantization efficacy.
This accomplishment exemplifies a pivotal shift, illustrating how the focus in AI optimization is extending beyond mere model compression algorithms to encompass the intricacies of data design and utilization.
Note AI’s successful participation in the hackathon not only aligns with this trend but also strengthens its partnership with NVIDIA.
Moreover, Nota AI is poised to further this collaboration across visual AI applications by incorporating NVIDIA’s video search and summarization tool, “VSS Blueprint,” into its vision-language model (VLM)-based real-time video analytics solution, Nota Vision Agent (NVA).
This integration facilitates the real-time detection and summarization of anomalous situations, thereby optimizing on-site response times.
Myungsu Chae, CEO of Nota AI, remarked, “This accolade underscores that AI optimization transcends algorithmic enhancement. It highlights the potential for innovation rooted in our design and application of purpose-fit data.
Building on our collaboration with NVIDIA, we aspire to advance data-centric AI optimization technologies and solutions with direct applicability in industrial environments.”
Source link: Prnewswire.com.
