Dr. Karthik V. Sarma, MD PhD FAMIA

Founder, AI in Mental Health Research Group, UCSF

Dr. Karthik V Sarma, MD PhD FAMIA is a physician and computational health scientist at the University of California, San Francisco, where he founded the UCSF AI in Mental Health Research Group. His work focuses on how large language models can be made safe and effective for psychiatric reasoning, diagnosis, and care delivery.

Dr. Sarma’s research spans the evaluation of AI systems on clinical tasks in psychiatry, methods for integrating expert clinical knowledge into AI models, and the effects of consumer AI chatbots on people living with mental illness, including early clinical reports of AI-associated psychosis. His work has been published in journals including Psychiatry Research, JMIR, and NPP–Digital Psychiatry and Neuroscience, and has been featured by outlets like The Wall Street Journal.

A computer scientist by training, Dr. Sarma is also co-founder and Chief Technology Officer of SimX, a virtual reality medical simulation company, where he led a research program supported by more than $20M in extramural funding. He is Trustee-Elect of the American Psychiatric Association, a delegate to the American Medical Association House of Delegates and co-chair of its Informatics Caucus, and a former member of the AMA Board of Trustees. He is a Fellow of the American Medical Informatics Association.

He received his M.D. and his Ph.D. in medical and imaging informatics from UCLA, and his B.S. in computer science from the California Institute of Technology.

Explore their work

Integrating Expert Knowledge into Large Language Models Improves Performance for Psychiatric Reasoning and Diagnosis

Psychiatry Research, 2026; 355:116844

A study showing that general-purpose language models overdiagnose psychiatric conditions when prompted directly, and that building in expert-derived diagnostic decision trees substantially improves their precision.

Responsible and innovative AI for mental health care: five priority themes

NPP—Digital Psychiatry and Neuroscience, 2026; 4(1):17

A consensus paper from an American College of Neuropsychopharmacology study group outlining five priorities for translating AI into clinical mental health care, including rigorous safety evaluation of patient-facing tools.

“You’re Not Crazy”: A Case of New-onset AI-associated Psychosis

Innovations in Clinical Neuroscience, 2025; 22(10-12):11

A case report documenting new-onset psychosis in the setting of immersive chatbot use, in which the chatbot validated and reinforced the patient’s delusional thinking.

Federated learning improves site performance in multicenter deep learning without data sharing

Journal of the American Medical Informatics Association, 2021; 28(6):1259-1264

A study across three academic institutions showing that a model trained with federated learning outperformed and generalized better than models trained at any single site, without sharing or pooling patient data.