Columbia University · New York
Medical AI Developer
I build machine learning systems that read medical images and clinical text — retinal scans, brain MRI, discharge summaries — and return something a person can actually act on.
Six studies below
About
Most people meet one half of me at a time. I'm on the pre-med track at Columbia, and I spend my research hours training generative models on brain scans. The two are the same ambition — I'd rather understand a disease well enough to write code that finds it than pick one side and specialise into a corner.
What that looks like in practice: at the Laboratory of AI & Biomedical Science I'm working on why Alzheimer's presents so differently between patients, using diffusion models to pull apart subtypes that a single diagnosis hides. Everything else on this page is me learning the craft that makes that possible — a tumour classifier, an eye-age model, a translator for discharge summaries my grandmother couldn't read.
Three places made me. I grew up in Belgium, spent time in the mountains of Boulder, my birthplace, and moved to New York in 2025 for Columbia. I speak English and French natively enough to argue in both, which turns out to be the useful threshold. Before any of this I co-founded a company that turned scrapped hot-air balloons into products — I'm most at home somewhere between the clinic and the terminal.
Six things I built, roughly newest first. Ongoing research is marked. Code is on GitHub for everything that has any.
Identifying biologically distinct subtypes of Alzheimer's disease from multimodal neuroimaging and genetic data. Extends the Smile-GAN / Surreal-GAN / Gene-SGAN line of work with diffusion-based generative models for more stable subtype clustering.
PyTorch · Diffusion models · Multimodal neuroimaging · Columbia University Irving Medical Center
Coming Soon
Rewrites discharge summaries, labs and radiology reports at a chosen reading level, and checks every number in the output against the source document.
React · LLM API · Prompt evaluation
GitHub
Predicts biological age by extracting microstructural biomarkers from retinal OCT images — the retina being the one place you can inspect neural tissue and microvasculature without cutting anything open.
PyTorch · Deep learning regression · Retinal OCT
GitHub
Detects brain tumours from MRI at 96% accuracy. Normalisation, augmentation and regularisation throughout, to keep the thing honest on scans it has never seen.
PyTorch · ResNet-18 · AdamW · Data augmentation
GitHub
Classifies autism spectrum disorder from responses to a standardised questionnaire. Built alongside a speech-language pathologist, so the output stays interpretable to the clinician who has to use it.
Python · scikit-learn · pandas
GitHub
Five-card draw video poker: full hand evaluation, betting and credit state, built as a desktop app in Java Swing.
Java · Swing · OOP
GitHubResearch, building things, and a few rooms I talked my way into. Newest first.
Aug 2019 — May 2025
Wavre, Belgium
CESS, Advanced Sciences & Mathematics: GPA 3.99. Alongside it: HarvardX CS50 AI, an advanced mathematics course at ULB, and a 57-page independent paper on where the limits of scientific investigation actually fall.
2023 — 2025
Model European Parliament · Kerkrade, Netherlands
Main submitter on position papers for both, and led the European Parliament's largest party through the 2024 convention. Two Best Delegation awards. Mostly it taught me to argue in front of people who disagree.
2023 — 2025
The Hague International Model United Nations (THIMUN) Foundation · The Hague, Netherlands
Main submitter on 2025 convention position papers.
2023 — 2025
TU Delft / QuTech · SONACA Aerospace · Université libre de Bruxelles
Microwave-to-optical quantum transducers at Quantum Delta NL. Aircraft and space structures at SONACA. Quantum optics and computational modelling in R and MATLAB at ULB.
2023 — 2025
École Internationale Le Verseau · Wavre
Started the school's robotics club from nothing and led it through building autonomous drones.
2023 — 2025
FlyHigh Enterprise · Belgium
Turned decommissioned hot-air balloon tarps into consumer products. Roughly €11,000 in profit, a slice of it to nonprofit partners, national press coverage, and second place out of 475 in the Belgian entrepreneurship competition.
2023 — 2025
École Internationale Le Verseau · Wavre
Promoted awareness of the European Union’s role and impact through schoolwide initiatives. Helped organize a fundraiser for the Restos du Cœur, a charity that provides hot meals to those in need, raising over €2,000 .
Aug 2025 — May 2029
Columbia University · New York
Rajasekharan Family Undergraduate Scholarship. Engineers Without Borders and Formula SAE on the side. Phi Gamma Delta fraternity.
Sep 2025 — Present
Columbia University Irving Medical Center · Laboratory of AI & Biomedical Science
Machine learning at the edge of neurodegenerative disease — finding biologically distinct subtypes of Alzheimer's in multimodal neuroimaging and genetic data. I design loss functions for clustering, disentanglement and latent priors, and take the results back to clinicians to check they mean anything.
Python · PyTorch · NumPy · Pandas
Honors & awards
Away from the screen
Long enough to get through La Campanella and Le Torrent, which mostly taught me that hard things yield to stubbornness spread over years rather than talent applied in bursts. It is the same lesson research keeps teaching me, only louder.
More performances