Accurate, accountable models from incomplete data.
What can a model infer about a system we can only partly observe, and how do we make its output trustworthy?
Transmission networks are always sampled incompletely. We train graph neural networks to impute the ties that surveys miss, recovering network structure with enough fidelity to support transmission modeling (Communications Medicine, 2026), and to find the overlapping communities that guide where prevention is placed.
The same emphasis on accountable models carries into public health operations. With India's national HIV program, we are developing interpretable, scenario-aware forecasts of antiretroviral therapy demand, built for the administrators who make supply decisions.
In medicine, we build AI systems for clinical decision support and medical education that keep each output linked to the guideline or primary evidence it rests on, so clinicians and faculty can check the reasoning.
New project, 2026
OLIVER, a robotic research interviewer
OLIVER (Omni-linked Intelligent Virtual Entity for Research) is a platform we are building for translational and clinical research. It pairs an embodied robotic interviewer with an evidence-linked agent harness that keeps each step of the model’s reasoning traceable.
The interviewer runs entirely on local hardware, with on-device speech, vision and language models, so it can conduct structured, REDCap-integrated interviews and network inventories without sending participants’ responses to outside services. It is designed as a middle ground between interviewer-administered surveys, which invite socially desirable answers, and self-administered ones, which depend on literacy and language.
Supported by a Next Generation Scholar Award from the Johns Hopkins Center for Innovative Medicine.
Dr. Clipman and Dr. Maunank Shah created PRISM, an AI-powered platform for medical education, and co-founded Equanimity Labs to develop it. PRISM carries the same emphasis on accountable AI into teaching: learners work through clinical cases in dialogue with an AI tutor that asks Socratic questions and gives feedback against reasoning rubrics, and faculty build the learning tools from their own course material.
Participation by Drs. Clipman and Shah in Equanimity Labs does not constitute or imply endorsement by the Johns Hopkins University or the Johns Hopkins Hospital and Health System.
