Hanoi, October 16, 2026 – FPT and NVIDIA have announced the expansion of Nemotron-Personas-Vietnam from 100,000 to 1 million records, a tenfold increase from the initial release. The dataset contains synthetic profiles of hypothetical Vietnamese individuals, providing researchers and developers with additional data to train, test, and evaluate artificial intelligence (AI) applications.
The dataset consists of AI-generated profiles representing hypothetical Vietnamese individuals. Each profile contains 21 data fields, including six demographic attributes and 15 contextual attributes, covering age, gender, education, occupation, place of residence, skills, and interests, along with descriptive passages about work, sports, travel, and food. All profiles are generated by AI and are not collected from real individuals.
Photo: FPT.To generate these synthetic profiles, the system uses statistics from two surveys conducted by the General Statistics Office of Vietnam (now the National Statistics Office under the Ministry of Finance): the 2024 Mid-Term Population and Housing Survey and the 2024 Vietnam Household Living Standards Survey.
The system uses these statistics to combine characteristics such as age, occupation, education, and place of residence in accordance with Vietnam's actual population distribution, creating hypothetical user profiles. AI then enriches the profiles with personality traits, skills, interests, and descriptive narratives.
During development, FPT contributed source data, computing infrastructure, research capabilities, and expertise in Vietnam's demographic, cultural, and social characteristics. FPT also participated in evaluating the dataset's quality. NVIDIA provided the methodology and tools for generating synthetic user profiles using AI.
The two companies first introduced the dataset in June 2026. Within four days of launch, Nemotron-Personas-Vietnam ranked among the Top 15 trending datasets on Hugging Face, a platform for sharing AI datasets and models. The dataset was made publicly available for researchers and developers to use under the applicable license terms.
An integrated AI system generates synthetic personas based on Vietnam's actual population distribution. Photo: huggingface.co.The initial release focused on six provinces and cities. In this expansion, FPT and NVIDIA have increased the number of records and broadened the dataset's coverage to represent more population groups, occupations, and living circumstances across different regions of Vietnam.
This enables developers to create additional scenarios for testing AI applications with diverse user groups. Developers can also expose AI systems to challenging or unusual situations to identify inaccurate or inappropriate responses.
Many AI datasets available globally may not fully reflect the characteristics and needs of Vietnamese users. Synthetic profiles allow developers to create testing scenarios that more closely resemble real-life situations in Vietnam.
For example, when developing an AI assistant for banking services, a technical team could simulate an elderly user who needs to check transaction history on a mobile phone. The team can submit this request to the AI system, assess whether the instructions are accurate and easy to understand, and make adjustments where necessary.
This testing approach can also reduce the need to use real customer profiles in certain situations.
The dataset can also support the testing of Agentic AI, a type of AI capable of performing multiple steps to fulfill user requests rather than simply answering questions.
For instance, when connected to appropriate tools and granted the necessary permissions, an AI assistant supporting administrative procedures could look up eligibility requirements, check the documents that need to be prepared, and assist users in completing forms.
Using synthetic profiles, developers can generate a wide range of requests to assess whether AI systems correctly understand tasks, execute each step accurately, and respond appropriately to different user groups.
Synthetic profiles can support AI testing in banking, healthcare, insurance, and public services. However, they do not replace the need to verify system accuracy and safety before deployment.
The expansion of Nemotron-Personas-Vietnam also contributes to the development of sovereign AI, providing Vietnam with additional data resources to independently research, build, and evaluate AI systems tailored to the characteristics of domestic users.
In particular, in highly regulated sectors such as finance, healthcare, insurance, and public services, the dataset can support the development and testing of specialized AI agents.
Through synthetic user profiles, developers can evaluate how AI handles requests under different circumstances, identify errors, and detect potential risks of biased responses or regulatory noncompliance before deploying systems in real-world environments.
Dataset information:
https://huggingface.co/datasets/nvidia/Nemotron-Personas-Vietnam