Nota AI Wins Grand Prize at NVIDIA Nemotron Hackathon With Data-Centric MoE Quantization - 한국면세뉴스

Takes first place in Track C and overall, underscoring a shift toward data‑centric AI optimization Nota AI, a Korean startup focused on AI model compression and optimization, has won the Grand Prize at the NVIDIA Nemotron Hackathon using synthetic data tailored for Mixture of Experts (MoE) quantization. The company took first place in Track C, which focused on synthetic data pipelines for SDG use cases, and was ranked top overall among 20 participating teams. The hackathon was part of “NVIDIA Nemotron Developer Days Seoul 2026,” held April 21–22 at d·camp Mapo in Seoul. NVIDIA said this was the first time its Nemotron Developer Days program, previously only at the GTC conference, was brought to Korea. The event combined technical masterclasses with a 48-hour hackathon focused on the Nemotron open model family and sovereign AI development. Participants competed across three tracks: agentic systems, domain-specific Nemotron models, and synthetic data pipelines. Final winners were announced on April 22. In its winning entry, Nota AI used an agent based on NVIDIA’s Nemotron 3 Super120B model to build a dataset optimized for MoE architectures. The company emphasized a data-centric approach, engineering the structure, quality, and task alignment of the synthetic data to enable MoE models to be quantized with minimal performance loss. This contrasts with conventional quantization work that focused mainly on formula- and algorithm-driven techniques. Beyond the hackathon, Nota AI is collaborating with NVIDIA in video analytics. Nota Vision Agent (NVA), based on vision-language models, integrates NVIDIA’s Video Search and Summarization (VSS) Blueprint to detect and summarize abnormal events in real-time from video streams. According to the company, this setup is already being applied to shorten response times in field operations. “This award shows that AI optimization does not stop at improving algorithms, but can also open new possibilities by designing and using data that is fit for purpose,” said Myungsu Chae, CEO of Nota AI. He added that the company plans to continue developing data-centric AI optimization technologies for real-world industrial deployments in collaboration with NVIDIA. By Min Byung-kwun, KDFN kdf＠kdfnews.com Source_ⓒNota AI