Nemotron-Personas-Brazil: Co-Designed Data for Sovereign AI (huggingface.co)

NVIDIA has launched Nemotron-Personas-Brazil, an open dataset comprising 6 million fully synthetic personas designed to enhance AI development for Brazil, a nation characterized by its linguistic and cultural diversity. This collection, grounded in official census and labor data from the Brazilian Institute of Geography and Statistics, reflects the demographic, geographic, and occupational distributions of Brazil's population. Unlike existing datasets, it offers resources that are culturally informed and commercially usable, thereby supporting local developers in building sovereign AI systems that respect regional identities. This significant release addresses a critical gap in AI training data, which has historically been centered around English and often fails to represent underrepresented populations. With over 1.4 billion tokens, 20 fields per record, and detailed persona attributes, such as language proficiency, skills, and interests, developers can create culturally-aware AI models. The dataset can be accessed through NVIDIA’s NeMo Data Designer, enabling tailored persona generation within AI frameworks. By prioritizing privacy—with no personally identifiable information included—and adhering to Brazil's data protection regulations, Nemotron-Personas-Brazil encourages ethically responsible synthetic data practices while fostering innovation in AI applications across Latin America.