Nemotron-Personas-Korea is a dataset of 6–7 million fully synthetic personas grounded in official Korean government statistics (KOSIS, Supreme Court, National Health Insurance Service), designed to help AI agents serve Korean users with culturally accurate language, regional context, and domain expertise. The dataset covers 26 fields including occupation, region, life stage, and persona type, and contains zero PII in compliance with Korea's PIPA. A step-by-step tutorial walks through loading the dataset, filtering personas by occupation or region, constructing a Korean-language system prompt, and deploying an agent via the NVIDIA API catalog, NIM, or NemoClaw. The post demonstrates how persona grounding changes agent responses from generic global guidance to Korea-specific public health advice in proper formal Korean (존댓말).
Table of contents
A Sovereign Dataset for South KoreaWhy This Matters for Autonomous AgentsTutorial: From Synthetic Persona to Sovereign AgentWhat Grounding ChangesCome Build with Us in SeoulSort: