Nemotron-Personas-Finance-Japan: A Synthetic Dataset Capturing Japan's Demographics and Financial Context

Elith and NVIDIA have introduced Nemotron-Personas-Finance-Japan, a synthetic dataset designed to support agentic AI development in Japan’s financial sector while addressing privacy and compliance constraints. The dataset extends the open-source Nemotron-Personas-Japan by incorporating Japan’s official statistics on demographics, household structures, and asset statuses, enabling realistic financial scenario simulations without using real customer data.

Developed using NVIDIA’s NeMo SDG-PGMs and NeMo Data Designer, the dataset was refined with input from major domestic financial institutions and industry groups, including Sompo Japan, Tokio Marine Holdings, and Mizuho Financial Group. It contains 9 million fully synthetic personas with 50 fields each, covering household finances and financial literacy, and is released under the CC BY 4.0 license for commercial and research use.

Nemotron-Personas-Finance-Japan is tailored for AI teams in Japan’s banking, insurance, and startup sectors, enabling applications such as customer-service agents, sales conversation evaluations, and user simulations in NeMo User Sim. The dataset’s scale and granularity—50% larger than its predecessor with 50 fields instead of 22—provide finer population slices for training and testing AI models in Japanese-language contexts.

The release complements Nemotron-PII-Japan, a companion dataset of 100,000 synthetic Japanese business documents with labeled personal information, designed for training data protection models. Together, these datasets offer a privacy-conscious foundation for developing culturally grounded AI systems in Japan, supporting both local developers and global teams targeting broader regional deployment.