We explore whether survival model performance in underrepresented high- and low-risk subgroups—regions of the prognostic spectrum where clinical decisions are most consequential—can be improved through targeted restructuring of the training data set. We propose a novel risk-stratified sampling method that addresses imbalances in prognostic subgroup density to support more reliable learning in underrepresented tail strata. Access the journal article here.

In high-stakes domains such as medical cost prediction and fraud detection, rare-event identification demands models that are simultaneously predictive and interpretable. Decision trees meet the interpretability requirement, yet under extreme class imbalance they typically collapse into shallow, majority-dominated structures that offer little value as auditable decision tools. We address this problem by formulating training data curation as a principled optimization problem that balances boundary support and maintaining geometric coverage.

We propose a multimodal multitask AI framework to predict 6 cardiovascular diseases using ECG time-series and electronic health records. We are currently designing the human-AI interface of our pilot model with practitioners at Hartford Healthcare, gaining insight into practical implementation challenges and clinical utility in routine practice. This project was awarded the bronze medal at the Yale Health AI Championship in May 2025.

In my final-year research project in Oxford, I developed a mixed-integer programming solver in Julia based on Clarabel.jl as the sub-solver.Together with Prof. Paul Goulart and Yuwen Chen, this work was published in the IEEE LCSS journal and CDC conference 2023. The code for the Clarabel.jl branch-and-bound wrapper can be found here.

We extended a human-robot dialogue framework to help end-users understand robot behaviour. I designed, developed, and evaluated human-robot question-and-answering simulations in ROS Gazebo. With Prof. Lars Kunze, we submitted to the ICRA2023 conference, entitled “Embodied Question Answering over Knowledge Graphs capturing Robot Observations and PDDL Plans” . Access my code here and here.

We designed the hardware, software as well as our business model for a service robot to help increase operational efficiency in drug dispensing at UK hospitals. The engineering and business design was awarded the 3YP Engineers in Business Competition Prize 2022 and selected as finalist in the EIBF Champion of Champions Competition 2022 at the Royal Academy of Engineering.