01SPEAKING · MEDIA
Speaking and media requests
Share the event or outlet, audience, format, deadline, and topic for keynotes, panels, grand rounds, podcasts, interviews, or expert comment.
GET IN TOUCH
Contact
Three focused forms route each request to the right review process. Applications to join the lab go through the Join page, not email.
02RESEARCH PARTNERSHIPS
Explore sponsored research or a collaborative grant
The lab collaborates with industry often. Its research has been supported by Microsoft, Google, and Amazon alongside NIH R01 funding, and with Gradient Health it released ReXGradient-160K, one of the largest public chest X-ray datasets. Propose a question aligned with one of the lab's research programs, with scope, timeline, and budget context.
This is preliminary intake. Any proposal or agreement is reviewed and authorized through Harvard.
03SUPPORT OUR WORK
Help ambitious medical AI research move forward
Gift support can advance open research, train early-career scientists, and provide the infrastructure needed to turn technical progress into clinical evidence. Philanthropy already powers signature efforts: the Biswas Family Foundation, in partnership with the Milken Institute, funds the lab's MAIDA Initiative to democratize global medical imaging data.
This form starts a conversation; it does not accept or process a gift. Contributions are coordinated through Harvard Medical School Alumni Affairs and Development.
Exploring a hospital or industry collaboration? See how collaborations with the lab work. Looking to join the lab as a postdoc, predoc, or Bootcamp participant? Start on the Join page.
SPEAKER KIT
For organizers: a ready-to-use talk title, abstract, and bio. To invite Pranav, use the speaking form above.
Talk: How Medical AI Earns Its License
Generalist medical AI can already describe a scan remarkably well. Describing an image is one capability. Localizing findings, reasoning in 3D, measuring honestly, and knowing when to doubt are four more, and today's models are weak at all of them. This talk follows that gap across three research programs at the Rajpurkar Lab: grounded reasoning and evaluation, where held-out benchmarks reveal what leaderboards conceal; clinical agents in simulated care, where a simulated hospital lets AI systems be tested against physicians before they ever touch a patient; and procedural learning and robotic assistance, where AI learns the physical structure of clinical work. The argument throughout: clinical AI should earn its license the way clinicians do, through graduated, supervised evaluation before deployment.
Short version
Medical AI describes scans well and falls short where clinical work gets hard: grounding findings in evidence, reasoning over a patient's course, and acting in the physical world. Drawing on work spanning generalist medical models, held-out evaluation, simulated hospitals, and surgical robotics, this talk argues that clinical AI should earn its license the way clinicians do, through graduated evaluation before it reaches patients.
Bio and headshot
Speaker bios in four lengths, plus a high-resolution headshot, are maintained at pranavrajpurkar.com/bio.