Publications
Papers, preprints, and tutorials from the lab on AI for psychiatry — statistical monitoring, multi-agent clinical systems, and the evaluation methods that test them.
2026
-
Context-Aware Hospitalization Forecasting Evaluations for Decision Support using LLMs
Compares direct LLM forecasting, classical time-series models, and a hybrid pipeline across 60 US counties, and finds LLMs work best as components inside structured models rather than as standalone forecasters.
-
Reliable Self-Harm Risk Screening via Adaptive Multi-Agent LLM Systems
A statistical framework for multi-agent pipelines structured as directed acyclic graphs, cutting the false positive rate by 40% against single-agent models without a matching rise in false negatives.
-
Reliability Auditing for Downstream LLM Tasks in Psychiatry: LLM-Generated Hospitalization Risk Scores
A structured perturbation analysis over four models and four prompt framings, showing that clinically insignificant changes to context shift predicted hospitalization risk.
-
Optimal Question Selection from a Large Question Bank for Clinical Field Recovery in Conversational Psychiatric Intake
A benchmark built from 655 clinician-written questions, showing that LLM-guided adaptive questioning recovers more clinical detail than random or fixed-form questioning, especially when patients are less forthcoming.
-
Decomposing Theory of Mind: How Emotional Processing Mediates ToM Abilities in LLMs
Finds that gains on belief-attribution tasks track patterns matching emotional processing rather than logical reasoning.
-
Auditing Generative AI Benchmarks with a Multi-Agent Compliance System
-
Constrained Process Maps for Multi-Agent Generative AI Workflows
Used to audit open-source benchmarks on mental health and self-harm topics.
-
Generative AI in Healthcare: Causality, Decision, and Real-world Case Study