A scientific story, and where it is going.

Past. I trained as an applied mathematician (B.Sc. with distinction; M.Sc. with honors), working on numerical methods for nonlinear dynamical systems — solver design, convergence, and stability for stiff and memory-bearing equations. That foundation became a data-science program at West Virginia University, where my Ph.D. in Applied Mathematics (2024) developed a data-driven method for inferring heterocellular networks in cancer: ensemble Bayesian causal-network inference from high-dimensional gene-expression data, combining deep learning with scalable graph algorithms.

Present. At the WVU Cancer Institute I moved fully into biomedical machine learning — generative models (VAEs, Wasserstein GANs), ensemble Bayesian network inference with MCMC, and cell-type deconvolution connected to community benchmarks. An NSF Research Fellowship in AI & ML for Digital Health (2023–2024) funded a scalable platform for causal-network inference, which became BaMANI (arXiv, 2025). As a Postdoctoral Associate at Duke (Biostatistics & Bioinformatics) I now build ML-integrated multiscale models — mechanistic simulators coupled with gradient-boosted trees, Gaussian-process surrogates, and physics-informed neural networks — and reproducible cross-dataset harmonization pipelines, contributing to the NIH/NIAID R01 (AI173333) Duke–Weill Cornell congenital-CMV collaboration.

Present & forward. Now, as Project Co-Investigator and Visiting Scholar at MIT Critical Data (2026–2027), I am building an independent research programme on the reliability and harm of perioperative and multimodal clinical AI: calibration and conformal uncertainty under distribution shift, subgroup and fairness auditing, consensus harm-scoring, and care-phenotype construction over multi-site EHRs (MIMIC-IV, MIMIC-CXR). Where the Duke–Weill Cornell project asks how maternal immunity protects against congenital CMV, my programme asks how we make clinical AI trustworthy enough to deploy in perioperative and critical-care settings — the methods are shared; the questions are my own.