Benjamin Herdeanu ↗
Causal discovery and dynamical systems.
Lead author · CausalDynamics ↗Join the lab · Heidelberg
Work on a question that is still open.
We are building AI that learns from first principles and adapts when the world changes. The hardest parts are unsolved. That is exactly why we are hiring.
TipPFN · CausalDynamics · BobPFN — proof points from the first year.
01The questions
Today’s models scale by memorising. We want models that scale by inferring.
That bet turns into concrete scientific problems. These are some of the ones nobody has solved yet — including us.
Causal discovery
Can we find what drives a system when the evidence is incomplete?
Start from CausalDynamics ↗Generalisation
How far can early warning travel to systems a model has never seen?
Start from TipPFN ↗Counterfactuals
How do we move from prediction to reliable what-if answers about interventions?
Read the approach ↗Synthetic priors
Which worlds should a model learn from, so that it can infer over the widest space of problems?
Start from TipBox ↗02What we offer
Responsibility for the science, from day one.
i.
Research ownership
Own a research problem end to end: from the synthetic prior to the model, the paper and the first real application with a design partner. You set the direction for your question.
ii.
Publication practice
We publish. Results go to peer review and arXiv; benchmarks and code go into the open so they can be checked and built on.
NeurIPS 2025arXivOpen code
iii.
Resources
€12M seed funding from UVC Partners, Entourage, HTGF and Mätch. Synthetic training data means the limiting factor is ideas, not data access.
Compute accessConference budget
iv.
Collaboration
Physics, mathematics, causal inference and machine learning in one team. Ideas get challenged across disciplines — and tested against real systems with partners.
03Your colleagues
Know whose thinking you’ll be challenging.
Yunus Sevinchan ↗
Complex systems and collective behaviour.
Co-lead author · TipPFN ↗Kai Ueltzhöffer ↗
Bayesian inference and the Free Energy Principle.
Co-lead author · TipPFN ↗
04How hiring works
A conversation about research, not a quiz.
Process to confirm
1
First conversation
What you have worked on, what you want to work on next.
2
Research discussion
Walk us through a piece of your work. We bring one of our open questions.
3
Meet the team
A day in Heidelberg with the people you would work with.
4
Decision
A clear answer, quickly — and the question you would own.
05Open roles
Find your question.
Role list to confirm
The next contribution
could be yours.
Bring a hypothesis, a result, a way of building.