Jean Technologies
Computers that understand humans.
Jean is building a foundation model for human behavior. Our models can predict what a person will say or do in any context.
We train on real human outcomes across billions of scenarios. Jean's models have learned to understand behavioral signals as a language of their own, and can be fed into downstream applications across a wide variety of tasks.
hu what the person has done/zu the representation learned from it/c context/a what they do next. The tasks compose this rather than reduce to it.
What reads off one representation
- Memory
- Keep what matters about a person and return it at the moment it is needed.
- Matching
- Connect people to people, and people to opportunities, on behavior rather than stated preference.
- Recommendation
- Surface what someone needs before they have gone looking for it.
- Personalization
- Software that adapts to the individual without being briefed each time.
- Simulation
- Rehearse a product, policy, or decision against a population before it meets the world.
- Forecasting
- Anticipate where a person is heading, from a health outcome to a churn risk, on the trajectory rather than a snapshot.
Reading
A summary of our survey in draft: three fields, one equation, and the corpus still to be built. The reading list behind it is public.
How Might We Teach Computers to Understand Us? →Research
Papers and notes on representation learning for people, retrieval, and the limits of a single embedding.
Research →Working together
We work with teams whose product depends on understanding the person using it. Palo Alto, California.
Contact →