How to Ground a Korean AI Agent in Real Demographics with Synthetic Personas

Nemotron-Personas-Korea fixes this. The dataset provides 6 million fully synthetic personas grounded in official statistics and seed data from the Korean Statistical Information Service (KOSIS), the Supreme Court of Korea, the National Health Insurance Service, and the Korea Rural Economic Institute. NAVER Cloud contributed seed data and domain expertise during design.

Every persona is demographically accurate but contains zero personally identifiable information (PII). It’s designed with Korea's Personal Information Protection Act (PIPA) in mind. South Korea is also one of the few countries to publish an official Synthetic Data Generation guide, establishing governance for grounding models with synthetic versions of sensitive data. This dataset follows that approach.

In this tutorial, we'll turn a synthetic persona into a deployed Korean agent — from filtering the dataset to inference — in about 20 minutes using hosted APIs.

| Attribute | Detail | |---|---| | Total personas | 7 million (1 million records × 7 personas each) | | Persona fields | 26 fields: 7 persona fields, 6 persona attribute fields, 12 demographic & geographic contextual fields, and 1 unique identifier | | Geographic coverage | All 17 Korean provinces, and 25 districts | | Names | ~209K unique names (118 surnames, ~21.4K given names) | | Occupations | 2K+ categories reflecting tech, manufacturing, public sector, etc. | | Persona types | Professional, family, sports, arts, travel, culinary, concise | | Life stages | Student, military service, employed, unemployed, retired | | Language | Natural Korean | | License | CC BY 4.0 |

Nemotron-Personas-Korea was generated using NeMo Data Designer, NVIDIA's open-source compound AI system for synthetic data. The pipeline pairs a Probabilistic Graphical Model (Apache-2.0) for statistical grounding with Gemma-4-31B for Korean-language narrative generation. Population data comes from KOSIS (2020–2026 releases); name distributions come from the Supreme Court of Korea.

Nemotron-Personas-Korea is the latest addition to the Nemotron-Personas Collection, which also covers the USA, Japan, India, Singapore (with AI Singapore), Brazil (with WideLabs), and France (with Pleias). If you're building a multilingual agent that serves Korean users alongside other markets, you can blend personas across countries in the same pipeline.

Most agents today are identity-blind. They follow instructions without any grounding in who they're serving. For example, an agent that books a Korean hospital appointment using US scheduling conventions, or addresses a 60-year-old patient in 반말 (“banmal,” informal language), doesn't just feel wrong. It fails.

Nemotron-Personas-Korea changes this by giving your agent a Korean operating context. Load a persona into the system prompt and the agent inherits that persona's region, occupation, communication norms, and domain expertise.

This works across any agent framework. Deploy with NemoClaw (NVIDIA's open-source reference stack for always-on agents running in NVIDIA OpenShell sandboxes, on anything from RTX PCs to DGX Spark), serve through NVIDIA NIM for production inference, or call the NVIDIA API directly. The persona layer is framework-agnostic, acting as a well-structured system prompt grounded in real Korean demographics.