Elith Inc. (headquartered in Bunkyo-ku, Tokyo; Representative Director, CEO & CTO: Koki Inoue; “Elith”), with the cooperation of Professor Yamashita of Chubu University, its technical advisor, took part in developing Nemotron Personas Finance Japan, an extension to the financial domain of the Japanese synthetic persona dataset “Nemotron-Personas-Japan” led by NVIDIA. Elith was responsible for implementation, building the generation pipeline, and large-scale data generation, and reflected insights and reviews from domestic financial institutions in the generation process and quality improvement. The dataset inherits the existing attributes and personal contexts that reflect Japan’s demographics and social and cultural background, and adds attributes and personas useful for AI development and user simulation in the financial domain. It has been available on Hugging Face since November 10, 2026.

■ Background
Financial institutions are increasingly using generative AI for customer service, product guidance, and internal operations. Before AI is built into operations, verification needs to assume a diverse range of customers who differ in age, family structure, assets, financial literacy, and more. However, actual customer data cannot be used freely because of personal information protection and confidentiality. In addition, creating customer profiles for verification one by one makes it difficult to ensure comprehensiveness and consistency with the real population composition.
“Nemotron-Personas-Japan,” released by NVIDIA in 2025, is an open dataset of synthetic Japanese personas based on official demographic and labor statistics. In addition to attributes such as age, occupation, and region of residence, it describes cultural background, skills, hobbies, and career aspirations in natural Japanese. The new Nemotron Personas Finance Japan extends this Japanese social context to the financial domain, supporting model training and evaluation as well as user simulation.
■ Elith’s role in this initiative
NVIDIA led the project as a whole, and Elith, a member of the NVIDIA Inception program for startups, was responsible for implementation, building the generation pipeline, and large-scale data generation. With the cooperation of Professor Yamashita of Chubu University, our technical advisor, we carried out the following work.
- Researching the statistics needed to design financial attributes and reflecting financial practice expertise
- Sampling attributes based on public statistics and building a generation pipeline that accounts for consistency between attributes
- Generating personal contexts and personas using NVIDIA NeMo Data Designer
- Large-scale data generation and quality evaluation (attribute consistency, distribution comparison with official statistics, human review)
- Review of interim deliverables by financial institutions, and reflection of feedback in design and quality improvement
■ Features of the dataset
1. Attribute design based on official statistics
To capture the sociodemographic and geographic diversity and complexity of Japan’s population, Nemotron Personas Finance Japan models relationships between attributes using probabilistic graphical models and samples attributes, based on the national census published by the Statistics Bureau of Japan, Ministry of Internal Affairs and Communications; Japanese surname and given name data provided by Myoji-Yurai.net; and survey and statistical data published by J-FLEC (Japan Financial Literacy and Education Corporation). This expresses a diverse range of people’s attributes while reflecting Japan’s demographics.
2. Added financial attributes
While inheriting the basic attributes and personal contexts of Nemotron-Personas-Japan, we added attributes useful for AI development in the financial domain, such as income and assets, to Nemotron Personas Finance Japan. We incorporated the expertise of domestic financial institutions so that spending, saving, investing, and financial decision-making can be expressed more realistically and diversely.
3. Reflecting the practical perspective of financial institutions
We received input on expected use cases and necessary attributes from the perspectives of different financial sectors, including banking, non-life insurance, and life insurance. Through review of interim deliverables, we also obtained feedback on the usefulness of the data and the appropriateness of its expressions. No company-specific customer data or confidential information was used, and expertise in financial practice was reflected in design and quality improvement.
4. Diverse personas reflecting the context of Japanese society
Like Nemotron-Personas-Japan, Nemotron Personas Finance Japan is built on 1 million samples, expanding the number of personas from 6 million to 9 million and total tokens from about 1.4 billion to about 1.5 billion. By combining attributes with natural Japanese descriptions, it expresses diverse financial situations and decisions in addition to Japan’s social and cultural background.
5. Synthetic data that does not use real customer data
This is synthetic data generated artificially from public statistics and other sources, and no person is linked to a real individual. No customer data or confidential information from financial institutions was used.
■ Expected use cases
We expect the dataset to be used as follows in AI development and user simulation in the financial domain.
- Synthesizing multi-turn conversations: Using diverse personas as customers to generate multi-turn exchanges such as consultations and customer service, for use as training and evaluation data for AI
- Marketing and market research simulation: Simulating reactions to product and service ideas and descriptions to organize hypotheses for research and measures
- Customer simulation for sales training: Use in role-playing that assumes customers with different backgrounds and financial situations
Results obtained using synthetic personas are intended for organizing hypotheses and preliminary verification, and do not replace actual customer research.
■ Marketing support solution using synthetic personas
Elith is developing a marketing support solution that uses this dataset. It simulates how diverse synthetic personas respond to measures such as products, services, advertising, and campaigns, visualizing trends and needs by target segment.
This supports hypothesis testing before actual measures are carried out and the consideration of more effective marketing strategies. We will begin offering this solution going forward and verify its usefulness in actual marketing operations through proof-of-concept trials with financial institutions.
■ Comments
Ryuta Shimogauchi, Director and CAIO, Elith Inc.
“To express the diverse personas required in the financial domain as synthetic data, not only data generation technology but also design grounded in actual operations is important. For this project, we reflected insights gained from NVIDIA and the financial industry in the generation process, aiming for a dataset that can be used across a wide range of AI development.”
Aser Blanco, Global Head of Banking Strategy, NVIDIA
“Financial institutions in Japan are rapidly building their own AI in order to manage sensitive data themselves and to offer services tailored to the needs of the Japanese market. Open datasets that reflect real-world Japan are essential to making that happen. NVIDIA developed Nemotron Personas Finance Japan, which reflects Japan’s language, culture, and financial practices, together with Elith, which has expertise in large-scale synthetic data generation and quality evaluation. With this dataset, Japanese financial institutions can generate cases from synthetic data and evaluate AI applications and services tailored to customer needs, without relying on actual customer data.”
■ Outlook
Using this dataset, Elith will advance efforts to use AI safely in the financial sector, such as evaluating the quality of consultation AI and customer service AI at financial institutions and verifying AI guardrails. By combining this with our expertise in evaluation and verification, including our AI evaluation platform GENFLUX, we will help financial institutions create an environment where they can check AI quality without relying on actual customer data.
■ Consultation on AI adoption and development
We accept inquiries at any time about AI adoption, development, and application to business operations. Please feel free to contact us even if your specific challenges or plans have not yet taken shape.
■ About Elith Inc.
Elith is a tech company that develops and deploys AI supporting critical operations in society and industry, such as finance, manufacturing, and healthcare, built on AI safety and security. Centered on GENFLUX, our AI safety platform, we support companies in visualizing and controlling generative AI use, protecting confidential information, evaluating AI quality, and establishing governance frameworks. Through advanced research and the development of enterprise AI products, we provide an environment where companies can safely adopt AI and use it in production operations while continuously evaluating and improving it.
Company name: Elith Inc.
Representative: Koki Inoue, Representative Director, CEO & CTO
Founded: December 2022
Headquarters: Frontier Hongo I 6-A, 2-27-17 Hongo, Bunkyo-ku, Tokyo, Japan
Business: Research, development, design, planning, education, sales, maintenance, and consulting related to AI
URL: https://www.elith.ai
■ Contact
Elith Inc. Public Relations
E-mail: pr-team@elith.ai
※ Company names, product names, and service names mentioned are trademarks or registered trademarks of their respective companies.

