Dara Rouholiman
I'm an ML Research Engineer at Stanford AIM Lab, where I build and evaluate machine learning models for clinical medicine — from novel deep learning architectures for time-series forecasting to large language model evaluation in perioperative settings. Our recent work showing how tool-augmented LLMs can reduce clinical calculation errors from 64% to below 5% was published in Nature's npj Digital Medicine.
I'm a physical chemist by training, which gave me a habit of thinking from first principles — not just what works, but why it works. I taught myself ML by building things: co-founding Telesphora (opioid overdose prediction, deployed with the Connecticut Department of Public Health), leading ML at COR (at-home blood analyzer with 1,000+ users, 3 patents), and serving as PI on a DoD SBIR grant. Along the way I've published 12 papers, filed 3 patents, and learned that the best health ML comes from sitting close to clinicians.
I also teach — I'm Lead Instructor for Stanford SASI's Healthcare Innovation Internship, where I run seminars on clinical data structures and the ethics of health AI for pre-med and medical students.
Recent News
-
15 Mar 2025:
📄 LLMs’ clinical calculations evaluation paper published in Nature's npj digital med
Our paper on large language models’ clinical calculations performance is published in Nature’s npj digital medicine! Our study showed how bad LLMs are at clinical calculations, so we had to build a set of task-specific clinical calculation tools and show how giving LLMs the task-specific tools–medical calculators–reduce their incorrect calculations to below 5% (from 64% without the tools).
-
05 Oct 2024:
📚 Lead Instructor, Stanford Technology in Healthcare Fall Internship
Co-designed a four-session virtual curriculum of advanced, graduate-level seminars with hands-on lab sessions on Clinical Data Structures and Algorithms for interns.
-
04 Mar 2024:
🎤 Poster Presentation at World Congress of Anesthesiology 2024
Presented our study on Open-Source Large Language Models in Anesthesia and Perioperative Medicine: ASA-Physical Status Evaluation. Our findings Underscore the significant potential of open-source Large Language Models like Mixtral in transforming anesthesia and perioperative medicine.