I'm a PhD student in Biomedical Informatics at Columbia University, working on benchmarking and evaluation of clinical AI systems.
My research focuses on understanding when machine learning and foundation models for healthcare actually work, and for what reasons, so that we can make meaningful methodological progress in Health AI. I'm particularly interested in evaluation, robustness, generalization, privacy, and reliable deployment of AI in healthcare.
- Clinical AI evaluation & benchmarking
- Foundation models for electronic health records
- Machine learning for healthcare
- Robustness, generalization & reliability
- Privacy & security of clinical AI
- Clinical large language models
I'm currently developing methods and infrastructure for evaluating clinical foundation models across diverse prediction tasks and healthcare datasets, with an emphasis on evaluating methods instead of models.
Recent projects include:
- 🩺 Clinical AI benchmarking: task generation, task selection, and evaluation pipelines for foundation models over longitudinal health records
- 📊 Foundation models for health: evaluating structured-EHR models across clinically relevant prediction tasks
- 🔐 Clinical LLM privacy: studying information leakage and inference attacks against models trained on sensitive data
- 🌐 Federated learning: methods for learning across heterogeneous healthcare institutions without centralizing their data
- 🧪 Clinical trial NLP: large-scale normalization and characterization of clinical trial outcomes using LLMs
Much of my current work uses the Medical Event Data Standard (MEDS) ecosystem to make clinical-AI experiments easier to reproduce and compare across models, tasks, and datasets.
I primarily work with Python, PyTorch, Hugging Face, Polars, Hydra, and scientific Python tools.
Before Columbia, I studied Computer Science at Mines Paris – PSL and Applied Mathematics / Machine Learning in the MVA master's program at ENS Paris-Saclay.
My earlier work spans machine learning, optimization, neurotechnology, scientific computing, and large-scale data analysis.