PHASE I
I built TTwin (Tumor Twin), a novel mathematical framework for representing cancer as an evolving system instead of a static prediction problem. Rather than predicting whether a patient will relapse, TTwin models the tumor as a dynamic distribution of cellular phenotypes that continuously changes under treatment. The biggest challenge was designing an entirely new mathematical representation that could realistically simulate tumor evolution, therapy interactions, and patient adaptation while remaining computationally tractable. I'm most proud that the framework introduces several original concepts—including measure-based tumor states, composable therapy operators, adaptive data assimilation, and counterfactual treatment simulation—that were designed from first principles rather than extending an existing AI model. Phase I focuses on establishing and validating this theoretical foundation on controlled synthetic experiments, providing the basis for later validation on real patient data.
