I am an orthopedic surgeon and AI researcher working at the intersection of medicine and machine learning. At Harvard Medical School's Department of Biomedical Informatics I work with Dr. Pranav Rajpurkar on AI for medical imaging, clinical decision support, and healthcare evaluation. After seven years and over a thousand knee surgeries, I build AI grounded in the realities of clinical practice — across 40+ publications, including Nature Medicine.
Grounded reasoning, segmentation, and report generation across X-ray, CT, ultrasound, and MRI. Building datasets and benchmarks that hold models accountable to clinical truth.
Static benchmarks miss how AI behaves in the wild. We build simulators and clinically grounded metrics so models are evaluated the way they are deployed.
Multi-agent diagnosis, voice agents for prehospital care, and the under-studied failure modes of chatty models — sycophancy, hallucination, premature closure.
Deployed deep-learning pipelines for alignment measurement, fracture detection, and outcome prediction — validated across institutions and now used in real clinical workflows.
A simulation framework for evaluating medical AI the way it will actually be used — dynamic, multi-turn, and grounded in clinical decision pathways. Co-first authored with Dr. Luyang Luo.
READ PAPER ↗
A deep-learning radiographic analysis of 17,365 knees showing that population alignment drifts toward constitutional varus as osteoarthritis severity (K&L grade) rises — challenging assumptions about "normal" alignment in arthritic knees.
READ PAPER ↗