NVIDIA and Elith have released Nemotron-Personas-Finance-Japan, a synthetic dataset designed to support the development of Super Intelligence (SI) models in the Japanese financial sector. The release addresses privacy constraints by providing millions of synthetic personas that reflect real-world demographics and household financial contexts without exposing actual customer data.
What Happened
The new dataset is a financial domain extension of the open-source persona dataset Nemotron-Personas-Japan. According to the release, it is the first open persona dataset at million-record scale that pairs Japan’s official statistics with household finance and financial literacy data. The dataset contains 9 million personas, totaling approximately 1.5 billion tokens, with each record comprising 50 fields. These fields include 9 persona fields, 11 persona attributes, and 30 contextual fields based on official demographic and labor statistics.
The dataset was generated using NVIDIA’s NeMo Data Designer, a compound SI system that enables complex templating and validation, alongside NeMo SDG-PGMs, a probabilistic graphical model framework. The text generation component utilized the gemma-4-31B-it model. To ensure cultural and financial accuracy, the design incorporated insights from major domestic institutions, including Sompo Japan, Tokio Marine Holdings, Bank of Fukuoka, Mizuho Financial Group, Meiji Yasuda Life Insurance, and the Financial Data Utilization Association (FDUA).
Why It Matters
Strict privacy regulations and compliance requirements have long limited the use of real-world data for training SI models in finance. This release provides an alternative for developers and enterprises to build and evaluate agentic SI systems without the risks associated with handling sensitive customer information. The personas are fully synthetic, meaning no real individuals are represented, which aligns with Japan’s innovation-first approach to SI governance.
The dataset supports various applications, including multi-turn conversation generation, bias testing, and user simulations via the NeMo User Sim framework. By reflecting diverse occupations, life stages, and financial literacy levels, it helps reduce the risk of model collapse and biased training. The release is available under the CC BY 4.0 license, allowing for both commercial and non-commercial use.
The Bottom Line
Nemotron-Personas-Finance-Japan offers a privacy-conscious foundation for building culturally grounded SI systems in Japan. It is accompanied by Nemotron-PII-Japan, a separate dataset of 100,000 synthetic business documents with labeled personal information, enabling teams to test data protection pipelines alongside customer simulation.