Research Focus & Program
Research Overview
As a physician-scientist and PhD researcher in Biomedical Informatics at Arizona State University, my work bridges clinical domain knowledge and machine learning research to build health AI systems that clinicians and patients can safely trust.
Trustworthy Clinical LLMs
Large language models offer unprecedented capabilities in processing clinical documentation, summarizing complex medical histories, and supporting decision-making. However, deployment in real-world clinical environments requires rigorous evaluation frameworks to detect hallucinations, measure calibration, and assess domain safety.
Key Research Focus Areas:
- Automated hallucination detection in clinical summaries and consultations.
- Context-aware LLMs for sensitive health data classification and privacy protection.
- Evaluation metrics aligned with real-world physician workflow requirements.
Evidence Grounding
For health AI output to be actionable in clinical care, every claim must be explicitly grounded in authoritative biomedical evidence, electronic health record (EHR) data, or peer-reviewed literature.
Key Research Focus Areas:
- Retrieval-Augmented Generation (RAG) verification for clinical practice guidelines.
- Auditability tools to trace AI outputs back to source EHR clinical notes.
- Computational platforms (e.g. EviTrace) for evidence attribution.
Interoperable Health Data & Granular Segmentation
Health Information Exchanges (HIEs) require robust data segmentation to protect sensitive patient records (such as substance use disorder data under 42 CFR Part 2) while ensuring interoperability across care settings.
Key Research Focus Areas:
- FHIR-based granular data segmentation (SHARES project, NIDA-funded).
- Standardized terminology mapping (SNOMED-CT, LOINC, RxNorm, ICD-10).
- Scalable deployment architectures for privacy-preserving health data exchange.