About us
A neo lab building
the next foundation model.
Abel is building a world-dynamics foundation model that learns how systems change, what drives them, and what happens when we act. Our research brings together temporal learning, causal intelligence and feedback from reality — toward Temporal Superintelligence.
What Abel AI Lab is
Abel is an AI research lab built around a mathematical view of the world: complex systems have states, dynamics and structure that can be learned. We are building a shared foundation for understanding those systems, rather than rebuilding a separate modeling stack for every domain.
Our name comes from Niels Henrik Abel and the mathematical tradition celebrated by the Abel Prize. Our mathematical core grew from longstanding relationships at Fudan’s Department of Mathematics.
The team brings together:
- Academic backgrounds at Fudan, Stanford, Arizona State University, UCLA, USC and Cornell Tech
- Research and engineering experience at Meta, Snap, Adobe, Microsoft and Microsoft Research
- Systematic investing and company building
Together, these paths connect dynamic-systems mathematics, foundation models, causal inference, agent research and large-scale engineering.
Our thesis
The world is recorded. It is not yet modeled. Today’s foundation models learn from what humanity has already written, photographed and coded. Real systems are states in motion: thousands of variables change together across time, respond to interventions and continually produce new outcomes.
The next foundation model must learn how the world changes. It must identify a system from its measurements and context, understand its state, learn its dynamics and structure, and reason about multiple possible futures — including what changes under action A, action B or no action at all.
Abel combines foundation models with causal intelligence. Structural evidence helps reveal which variables and relationships drive change, while temporal learning models how those systems evolve across time, scale and regime.
Mission
Our mission is to build the foundation model that makes the changing world understandable, computable, and ultimately actionable.
Where our current work sits
Our technical route brings four capabilities into one research agenda: dynamic-systems mathematics, foundation-model learning, causal structure and engineering at the scale of real data. The aim is not simply to connect tools, but to form a consistent understanding of the same system.
Our current work develops complementary parts of that foundation. Each addresses a different question: how a system evolves, what drives it, and how observed outcomes can improve the next model.
Simons
A unified foundation-model framework for time-series and tabular data, supporting forecasting and other predictive tasks. It develops the temporal learning capability needed to generalize across heterogeneous systems.
Andrey
A very fast causal discovery package for Python. It explores large variable spaces to identify candidate structure, drivers and competing explanations — structural evidence for what a model should pay attention to.
RSI Loop
The learning-loop direction connects predictions with observed outcomes, so the difference can become evidence for evaluation, new training data and model improvement.
Long-term vision and values
Our long-term vision is the next foundation model: a model that can maintain an understanding of a changing system, reason about the consequences of actions and learn from what actually happens. We want a shared foundation across markets, energy, industry, biology and science — not another isolated model for each domain.
Start from structure, end with evidence
How we work starts with mathematical structure and ends with evidence. We ask what changes, what drives it and which relationships matter. We test candidate explanations against real observations, and use the gap between predictions and outcomes to improve the next model.
Uncertainty is part of the model
Uncertainty is part of the model, not something to hide. Where evidence is insufficient, the goal is to identify what further measurements or experiments are needed, or defer to human judgment. We distinguish research directions from demonstrated capabilities, and long-term plans from current results.