I'm currently a PhD researcher in Artificial Intelligence and Machine Learning in Biomedicine at Imperial College London. My research focuses on deep learning, continuous-time modelling, uncertainty quantification, and machine learning for longitudinal biomedical data. Previously, I was a Machine Learning Researcher at Stanford University, where I developed and evaluated machine learning models on large-scale biomedical cohorts.

I've worked on several machine learning and biomedical research projects, including:

• SynLV: A Benchmark for Decision-Time Incompleteness in Longitudinal Survival Prediction — NeurIPS Datasets & Benchmarks, under review (2026). A synthetic benchmark for evaluating longitudinal survival models when information available at decision time is incomplete.

• EPC Mirage: Grid Decarbonisation can Overstate Retrofit Progress in Carbon-weighted Building Energy Ratings — Nature Energy, under review (2026). Investigates how carbon-weighted building energy ratings can create apparent efficiency improvements without corresponding physical retrofit.

• Evaluating Molecular Disparities in Breast Cancer Leveraging Machine Learning — Journal of Clinical Oncology, 2024. Machine learning analysis of molecular disparities and heterogeneity in breast cancer.

• Assessing Tree-Based Phenotype Prediction on the UK Biobank — IEEE BIBM, 2023. Evaluation of tree-based machine learning approaches for high-dimensional phenotype prediction.

I've developed a dataset for stress-testing robustness on longitudinal and population-scale biomedical models:

• SynLV — A synthetic benchmark for studying decision-time incompleteness in longitudinal survival prediction, with controllable longitudinal dynamics, observation processes, and missingness mechanisms.

I've worked on several machine learning and biomedical data science projects featured on GitHub, including:

• Polybitrage – Built a low-latency arbitrage engine for Polymarket and Kalshi, targeting cross-market pricing inefficiencies. Implemented N-leg arbitrage detection, depth-aware position sizing, and latency-sensitive execution with fill-or-kill orders and risk limits. Achieved sub-100 µs scan latency using Python, Rust, and Cython, with end-to-end latency monitoring via Prometheus/Grafana.

• COUGHVID - Deep learning models for detecting upper respiratory tract infections from cough audio. Uses mel-spectrogram representations with CNN and LSTM architectures, with a focus on robustness to background noise and recording-device variability.

• CKD Patient Profiling - Proteomics-based patient profiling pipeline for chronic kidney disease. Uses UMAP and Gaussian Mixture Models for clustering, Random Forest and LASSO for feature selection, and survival analysis for clinical outcome modelling.

• Phenotype Prediction - Machine learning pipeline for high-dimensional phenotype prediction using UK Biobank data. Includes XGBoost, LightGBM, CatBoost, feature selection, and SHAP-based model interpretation.

• ACDS - Attitude Control and Determination System for a CubeSat, in portable C. Implements B-dot magnetorquer detumbling and a 3-axis quaternion PID pointing controller, with IGRF-13 geomagnetic-field and SGP4 orbit-propagation models for attitude determination.