FPT Corporation and NVIDIA expanded their open-source Nemotron-Personas-Vietnam dataset to 1 million synthetic profiles on 16 October 2026, according to a corporate disclosure distributed through PR Newswire. The release marks a 10-fold expansion from the initial version published in June 2026, which covered a narrower sample of major cities. The updated dataset is published openly on Hugging Face under a permissive licence compatible with NVIDIA NeMo libraries.

The synthetic data uses statistical anchors from Vietnam's General Statistics Office, specifically the 2024 Population and Housing Census and the 2024 Vietnam Household Living Standards Survey. Additional data modeling came from FPT Smart Cloud and the Quantum AI and Cyber Security Institute. The dataset contains 21 total fields, dividing into 6 core persona attributes and 15 contextual demographic variables, alongside 13,000 unique full-name combinations and 20 occupational categories drawn from official labor statistics.

Built on NVIDIA's Nemotron methodology, the project pairs national demographic distributions with open-weight language models to generate synthetic narrative profiles without collecting personally identifiable information. The 10-fold scale increase is designed to support fine-grained population segmentation, such as testing enterprise agents on rural agricultural workers or urban retirees, while reducing model collapse across synthetic training pipelines.

The release arrives as domestic enterprises face tightening technical demands for sovereign AI systems under the national digital development strategy. While global foundational models handle Vietnamese text translation, enterprise agent deployments in banking, healthcare, and public administration require accurate local cultural context, income distributions, and administrative categorizations to prevent compliance failures.