About
A digital world made of people, not numbers.
Microworld is an experiment in modeling collective behavior with synthetic people. Browse a million Korean personas in a force-directed graph, then talk to any one of them to find out what they would actually do — and why.
Build the digital world with AI-generated personas
Every node in Microworld is a synthetic person from nvidia/Nemotron-Personas — grounded in real demographic and geographic distributions of the place they live. Each one carries an age, occupation, and region alongside a multi-paragraph narrative covering work, family, hobbies, and how they spend a typical day.
The graph clusters them by semantic similarity over their full text, embedded with Qwen3, so neighbors are people who share context or interests, not just a postal code. One million Koreans live here today; seven countries are on the roadmap.
Simulate any decision
Click a node and chat with that persona. They reply in character, grounded in their full background — and they keep their voice across follow-up questions. Use it as a one-person focus group, a sounding board for product copy, or a way to pressure-test a pitch against someone whose situation differs from yours.
Whether you're shipping a policy, a feature, or a marketing line, you can rehearse it against the people it touches before it touches them.
Understand why, instead of predicting what
Statistical models tell you that 62% of 30-something Seoul renters disliked a feature. They can't tell you what made the third one stop using it.
Microworld inverts that. Instead of fitting a curve over a population, you talk to the population — one person at a time, with their full backstory in scope. The output isn't a confidence interval; it's a chain of human reasoning you can read and argue with.
Try it
Start from a random persona, expand their neighborhood, talk to anyone who catches your eye. Or describe yourself and find your closest twin in the corpus.