Aashna Shah, PhD
Berkowitz Postdoctoral Research Fellow · Harvard Medical School
I am a Berkowitz Postdoctoral Research Fellow in the Department of Biomedical Informatics at Harvard Medical School, working with Arjun (Raj) Manrai.
I work on AI for clinical decision making, asking what "normal" means for each patient. Using longitudinal electronic health records, I build methods that personalize population-based reference ranges and risk scores, so that disease can be caught earlier, diagnosed more accurately, and care can shift toward prevention.
Education
Research
Learning Representations of Patient Baselines
Learning what “normal” looks like for each patient from longitudinal laboratory histories and population-scale data, so that meaningful change can be detected earlier. I am especially interested in sequence models that balance a patient’s own baseline against population priors.
Related publications
Fairness and Robustness in Clinical Algorithms
Examining how demographic adjustments and model shortcuts shape clinical predictions, and replacing group-level proxies with the individual measurements that actually drive risk. I am especially interested in race-based clinical equations and the robustness of vision-language models across patient populations.
Related publications
Agentic Systems for Clinical Reasoning and Evaluation
Building AI agents that reason through diagnostic workups and recommend next steps in care, and developing ways to evaluate them against physician judgment at scale. I am especially interested in laboratory test recommendation and benchmarking on real-world health records.
Related publicationsFeatured Publications
† equal contribution · See the full publication list on Google Scholar
Featured Talks & Posters
