At the developer event “NVIDIA Agentic AI Day,” hosted by NVIDIA in Tokyo on November 10, NVIDIA announced two synthetic datasets, “Nemotron Personas Finance Japan” and “Nemotron PII Japan,” to support Japanese-language AI development. A finance-domain expansion of the open-source persona dataset “Nemotron Personas Japan,” it combines the dataset with data on Japanese official statistics, household financial circumstances, and financial literacy. By avoiding the privacy risks associated with using real data and reproducing more diverse and realistic financial scenarios, it is expected to support agentic AI development in the financial sector.

NVIDIA Announces Two Synthetic Datasets Supporting Agentic AI Development in Japan - NVIDIA | Japan Blog
https://blogs.nvidia.co.jp/blog/nvidia-agentic-ai-day-japan-2026-two-japanese-dataset/?ncid=so-twit-875483




The two datasets were released as part of “NVIDIA Nemotron,” the open model and dataset family for agentic AI. Both are synthetic datasets capable of reproducing the required attributes and document structures without using information from real individuals, enabling AI training that accurately reflects Japan’s language, culture, demographics, and industry-specific conditions.

In 2025, NVIDIA released “Nemotron Personas Japan” (V1), an open dataset of six million synthetic Japanese personas. The newly introduced Nemotron Personas Finance Japan retains V1’s basic attributes and individual contexts while adding attributes and personas useful for AI development and user simulation in the financial domain.

Nemotron Personas Finance Japan is built on a foundation of one million samples, expanding the number of personas from six million in V1 to nine million and the total token count from approximately 1.4 billion to approximately 1.5 billion. Every persona is artificially generated and is not linked to any real individual, making the dataset suitable for synthesizing multi-turn conversations, simulating marketing and market research, and simulating customers for sales training.

Nemotron-Personas-Finance-Japan: A synthetic persona dataset capturing Japan’s population distribution and financial context
https://huggingface.co/blog/nvidia/nemotron-personas-finance-japan-ja



The overall project was led by NVIDIA, while Elith, a Japanese AI startup and member of NVIDIA’s Inception program, handled implementation, development of the generation pipeline, and large-scale data generation. Takashi Shimogakuchi, Elith’s director and Chief AI Officer, stated: “To represent the diverse personas required in the financial sector as synthetic data, it is important not only to have data generation technology, but also to incorporate design based on actual business operations. This time, we reflected the insights gained from NVIDIA and stakeholders across the financial industry in the generation process, with the aim of creating a dataset that can be used for a wide range of AI development.” The project also reportedly received support from major Japanese financial institutions to enhance its applicability in financial operations.

In addition, another dataset, “Nemotron PII Japan,” was released simultaneously. “PII” refers to information that can be used to identify an individual. The dataset contains 43 types of documents, including medical certificates and application forms, covering eight fields: healthcare, finance, government administration, employment, education, law, IT/security, and e-commerce, and includes PII such as names, addresses, email addresses, and various identification numbers. No real personal information is used here either, enabling additional training, evaluation, and improvement of models for PII detection and masking without privacy concerns.

Nemotron PII Japan: An open Japanese PII dataset for privacy protection
https://huggingface.co/blog/nvidia/nemotron-pii-japan-ja


“Nemotron Personas Finance Japan” and “Nemotron PII Japan” are available on Hugging Face.