NVIDIA and Elith released Nemotron-Personas-Finance-Japan on November 9, 2026, opening nine million synthetic Japanese personas for commercial and non-commercial AI development. The dataset is designed to help banks, insurers and developers test customer-facing systems without using real customer records.
Licensed under CC BY 4.0, the release adds household finances, money behavior and financial literacy to demographic profiles. NVIDIA describes it, to its knowledge, as the first open persona dataset at million-record scale to combine Japan’s official statistics with household-finance and financial-literacy data.
From demographic profiles to financial lives
The new dataset extends Nemotron-Personas-Japan, whose September 23, 2025 version held six million personas across 22 fields but excluded finance-specific personas. The new edition contains one million Japanese-language records, with nine personas each and 50 fields per record.
Its coverage spans all 47 prefectures and more than 1,600 occupation categories. Three of the 20 persona-related fields separately describe financial habits, decision-making behavior and financial concerns, giving each profile more detail than basic age, location or employment information.
The remaining 30 contextual fields cover personal income, household assets, housing and mortgage status, consumption propensity and financial-literacy scores. They also include internet and smartphone use, household structure and self-rated health. Together, these details can seed multi-turn Japanese dialogue that reflects different household situations.
Examples include a regional pensioner asking about asset succession and an urban dual-income household reviewing education savings and a mortgage. Those situations give developers a way to test whether a customer-facing system responds consistently when financial questions connect to age, location, housing and family circumstances.
Statistics behind the synthetic households
The personas are synthetic, but their household-asset distributions draw on the Statistics Bureau’s 2019 National Survey of Family Income, Consumption and Wealth. The dataset card’s analysis reports that the share of households with at least 20 million yen in financial assets rises from about 10% among people aged 25–29 to about 30% among those aged 60–64.
The analysis also reports that assets remain relatively high through people’s 60s and 70s, even as personal income falls around retirement. That pattern matters for systems that need to handle questions about savings, pensions, inheritance or financial planning without treating income as the only measure of a household’s situation.
The synthetic personas cover adults aged 18 and older. The release also draws on J-FLEC’s survey and statistical data, along with J-FLEC’s official index, as named seed inputs for the dataset.
What developers can test
Nemotron-Personas-Finance-Japan is intended for customer-service agents, sales and marketing conversation tests, NeMo User Sim simulations and red-teaming. Developers can also use it to check whether systems behave consistently across rural and urban locations, ages, education levels and economic circumstances.
That range creates a practical testing ground for financial AI. A system might handle a mortgage question well for one household but struggle when the same question comes from a pensioner, a family with education costs or someone living outside a major urban area. Synthetic personas let teams run those comparisons without exposing real customer information.
The generation stack combines NeMo Data Designer, the NeMo SDG-PGMs probabilistic-graphical-model framework and the Apache-2.0-licensed gemma-4-31B-it model for Japanese text. Data Designer supplies structured outputs, validation and automated retries, helping the records follow the dataset’s required format.
Sompo Japan, Tokio Marine Holdings, Bank of Fukuoka, Mizuho Financial Group, Meiji Yasuda Life Insurance and the Financial Data Utilization Association provided feedback during the design process. Their involvement connects the dataset to the kinds of financial services and customer interactions it aims to support.
For developers, the main value is not simply the nine-million-person count. It is the combination of Japanese-language dialogue, financial behavior, household assets, regional coverage and demographic detail in an openly licensed dataset. That gives teams a larger setting for testing AI systems before those systems face real customer records.
Based on
