A compound AI approach to Japanese personas grounded in real-world distributions
Open Data Breaking Down Privacy Barriers to Pave the Way for AI in Finance
In the financial sector, the utilization of real-world data faces significant constraints due to strict privacy regulations, compliance requirements, and the need for advanced domain expertise. In this context, the development of agentic AI in the financial domain requires synthetic data capable of avoiding the risks of using real data while reflecting demographics, household structures, and asset statuses to reproduce diverse, realistic financial scenarios.
To address these challenges, Elith and NVIDIA have released Nemotron-Personas-Finance-Japan, a financial domain extension of the open-source persona dataset "Nemotron-Personas-Japan." To our knowledge, it is the first open persona dataset at million-record scale that pairs Japan’s official statistics with household finance and financial literacy data. Built upon official government statistics, this dataset was developed using NeMo SDG-PGMs, a probabilistic graphical model framework for synthetic data generation (SDG), and NeMo Data Designer, an enterprise synthetic data generation system. Furthermore, to enhance the practical utility of the dataset in financial use cases, it was designed by incorporating insights and feedback provided by leading domestic financial institutions and industry associations, including Sompo Japan, Tokio Marine Holdings, Bank of Fukuoka, Mizuho Financial Group, Meiji Yasuda Life Insurance, and the Financial Data Utilization Association (FDUA).
This dataset expands the environment where Japanese developers and enterprises can develop and evaluate AI tailored to local and industry requirements, without using real customer data. Released under the CC BY 4.0 license, it is available for commercial use. From building conversational agents to customer simulation and model evaluation, it supports practical financial AI development. The Nemotron-Personas collection, including Nemotron-Personas-Finance-Japan, is designed to seamlessly integrate with the user simulation framework NeMo User Sim, enabling instant execution of simulations drawn from diverse distributions consistent with the real world.
What’s in the Dataset?
- 9M personas total (1M records x 9 personas each), written in natural Japanese
- 50 fields per record: 9 persona fields, 11 persona attributes, and 30 contextual fields based on official demographic and labor statistics
- ~1.5B tokens total, including ~750M persona tokens
- ~950k unique names
- 1600+ occupation categories reflecting Japan's workforce
- Comprehensive coverage across demographic, regional, asset distribution, and personality trait axes
- Diverse persona types: Occupations, sports, arts, travel, cooking, finance, financial behavior, and financial concerns
- Natural language persona attributes: Cultural background, skills and expertise, career goals and aspirations, hobbies and interests, household/parental/financial background, attitudes toward money, and financial planning
- Available for both commercial and non-commercial use under the CC BY 4.0 license
How We Built It
Data Generation Pipeline
It was built using NVIDIA’s framework for synthetic data generation, NeMo Data Designer. This compound AI system enables complex Jinja templating, Pydantic validation, structured outputs, automated retries, and support for multiple generation backends. This is the essential toolset required for generating such large-scale synthetic datasets. In addition, the following models were utilized:
- NeMo SDG-PGMs (Apache-2.0): A probabilistic graphical model framework for statistically grounded data generation
- gemma-4-31B-it (Apache-2.0): For Japanese text generation
Enhanced Cultural Context
Nemotron-Personas-Finance-Japan was designed to align with Japan's official demographic and labor statistics while taking into account key considerations essential for AI training:
- Education: Where national statistics broadly aggregate degree levels, we introduced finer granularity to enable the model to reflect diverse educational pathways.
- Occupation: We incorporated additional categories (such as business owners and specialized professions) to broaden the spectrum of occupations used for training.
- Life Stages: We modeled scenarios that rarely surface prominently in statistics, such as students, retirees, and periods of unemployment, to represent more realistic personas.
- Household Finances: In addition to individual income and spending propensity, we incorporated household-level financial and real assets to replicate diverse economic backgrounds.
- Digital & Financial Literacy: By including internet and smartphone usage along with financial literacy, we enabled the model to precisely capture consumer profiles in financial use cases.
Private By Design
Every persona is fully synthetic. No real individuals are represented, and the dataset is designed to reduce privacy risk. Demographics and household finances follow official statistics and public survey data, and names follow real Japanese naming patterns. Personas are built from these distributions, not from records about real people.
Who This Data Is For
Nemotron-Personas-Finance-Japan is built first for teams creating financial AI in Japan, at banks and insurers and at the startups and integrators that serve them. Each persona carries household finances and financial literacy alongside demographics. Teams can use them to build customer-service agents, evaluate sales conversations/marketing initiatives, run user simulations in NeMo User Sim, and red-team customer-facing AI.
The release also serves anyone building for Japanese users. It holds 9M personas, 50% more than Nemotron-Personas-Japan, across 50 fields instead of 22. That gives any Japanese-language project finer slices of the population to train and test on.
Practical AI Applications
The synthetic personas in this dataset can be leveraged for:
- Multi-turn conversation: Using personas as "seeds" to generate human-like dialogue datasets.
- Domain-specific AI development: Generating datasets to build AI assistants tailored to realistic scenarios in financial use cases.
- Bias testing and fairness: Evaluating how models and AI agent systems perform across rural and urban divides, various age groups, educational backgrounds, and economic circumstances to check whether AI performs consistently across Japanese society.
Why It Matters
Access to diverse, high-quality training data that reflects real-world populations has long been a challenge in AI development. Enterprise AI development has predominantly relied on private data, creating barriers for researchers, startups, and particularly AI developers in regions with limited available data.
- Data Diversity: Reflects the full spectrum of Japan's population, which helps reduce biased training and the risk of model collapse.
- Cultural Authenticity: Reduces reliance on Western-centric datasets and supports the development of Sovereign AI systems.
- Privacy and Governance: Aligned with Japan’s innovation-first approach to AI governance. Every persona is synthetic, so no customer data is needed to build or test with it.
By releasing Nemotron-Personas-Finance-Japan under CC BY 4.0, we provide access to enterprise-grade, high-quality synthetic data, empowering anyone to build culturally grounded AI systems with lower cost and less privacy risk.
Pairs with Nemotron-PII-Japan
Nemotron-PII-Japan launches the same day. It holds 100,000 synthetic Japanese business documents with labeled personal information, for training and testing detection and redaction models. Its documents are built on Nemotron-Personas-Japan. Use Nemotron-Personas-Finance-Japan to simulate customers, and Nemotron-PII-Japan to test how your pipeline protects their data.
Start Building with Nemotron-Personas-Finance-Japan
You can download this dataset using the command below to help develop AI that genuinely understands Japanese culture and language.
from datasets import load_dataset
ds = load_dataset("nvidia/Nemotron-Personas-Finance-Japan")
Use cases for building production-ready applications:
- Using personas as seeds for dialogue generation
- Fine-tuning models with culturally grounded data
- Building personalized engines that reflect Japan's overall demographic makeup
- Developing domain-specific AI agents equipped with Japanese context
From teams building AI in Japan to global developers aiming for broader regional deployment, Nemotron-Personas-Finance-Japan dataset provides the authentic, privacy-conscious foundation required for your applications.
Instagram Engineering
Shopify Engineering