About

I work on the formal structure of learning in AI systems — specifically on how agents develop structured, compositionally valid representations of domain knowledge, and how that development shapes their capacity for robust transfer and cross-domain reasoning.

My current focus is the H-Bar Model, a theoretical framework that formalizes the distinction between parametric depth (δ) and schema coherence (σ) as independently necessary dimensions of agent capability. The framework draws directly on my work in Physics-Informed Residual Learning (PIRL), where the contrast between physics-structured and unstructured neural architectures makes the cost of depth without schema coherence concrete and measurable.

[Affiliation and programme details. Update before publishing.]