Social simulators,
powered by language model agents.
Yew AI builds multi-agent simulations of social phenomena — using LLMs to populate virtual societies that researchers, educators, and policymakers can probe before deploying ideas in the real world.
MissionThe world is too complex to test in.
Climate policy, social media regulation, school curricula, urban planning — every important social intervention is also a giant uncontrolled experiment, performed live on millions of people.
Yew builds the wind tunnel for ideas about people. We use LLM-powered agents — not as products, but as a substrate for computational social science — so that ideas can be tested in silico, with humility, before they reach the world.
What we makeOpen tools, open research.
- Simulation frameworks Multi-agent libraries with realistic conversation, memory, and group dynamics — designed for reproducibility.
- Replication studies Re-running classic social science experiments inside language model worlds, and reporting where they break.
- Teaching artifacts Course-ready notebooks for sociology, economics, and political science classrooms.
- Open weights & data Everything we train and collect lives on GitHub and Hugging Face, under permissive licenses.
We add: and some can teach you what you didn't know to ask.
StatusWhat's live now
Yew AI is in its earliest days. There's no product yet, no funding round, no team page — just a name, a domain, and a thesis we believe in. If that sounds interesting, follow along on GitHub or reach out.
Who's behind itFounder
Yew AI was founded in 2026 by Shan Shan, an independent researcher working at the intersection of language models and the social sciences. Reach out if you'd like to collaborate, contribute, or simply argue about whether agent-based models actually predict anything.
Want to build the social wind tunnel?
We're looking for collaborators in computational sociology, economics, agent-based modeling, and LLM tooling.