Unifying emergent and designed structure.
We research how AI learns, remembers and acts, and how deliberate structure can shape it.

Vierla · AI research in progress
Where learning meets design.
- 01Representing a task
What does a representation make possible? We study the patterns it helps a system learn, and what it leaves out.
- 02Connecting capabilities
When parts interact, what new behaviour becomes possible? We study the interfaces that shape those relationships.
- 03Choosing constraints
Which boundaries help learning without prescribing its answer? We ask where to place them, and what they exclude.
- 04Meeting the world
What can a system observe and change? Its interface helps define the problem it learns.
- 05Remembering what matters
Useful memory changes with context. We ask what should persist as the world changes.
- 06Testing the explanation
Change the structure, compare the behaviour. An experiment has to be able to challenge the idea.
An interactive aside
A few signals, many possibilities.
Each tile reads something happening on this page. Change a view or rearrange the tiles.
Drag a signal to move it. Press one to change its view.
Our approach
A question before an architecture.
- 01
- 02
- 03
- 04
- 05
- 06
We explore new architectures, including world models: systems that learn how an environment changes. Our questions span representation, memory, prediction and coordination.
Read the philosophyStructure across scales · illustrated
From parts to relationships.
- PartBegin with a part
A simple part gives us a place to examine an assumption.
- ArrangementChange the arrangement
The same parts can behave differently when the relationships between them change.
- ComparisonVary one thing
Replacing a part makes a comparison possible. What changes, and what remains?
- CompositionStudy the next scale
A system can become a part of something larger. Which properties carry across?
- InteractionLook between systems
Connections introduce coordination, dependence and new sources of uncertainty.
- PerturbationTest what holds
Remove a connection and observe what changes. This illustration asks a question about dependence; an experiment would need to test it.
Research in progress
What we’re working on.
Our work spans representation, memory, prediction and coordination. We’ll share research and applications as they develop.
Have a problem AI should solve? Tell us what a useful outcome would look like.
Occasional notes on our research and what we’re learning.