Geometric Interpretability for Model Reliability

I develop geometric methods for measuring what happens inside AI models - how representations form, move, and can be measured - and apply them to interpretability, safety, and the reliability of the latent models the field is converging on.

The methods are peer-reviewed, open-source, and have been independently cited and extended by another research group. Underneath them sits FRESH, the geometric framework for self-models that generates the predictions these instruments test.