Research Projects

Human-Centered AI in Healthcare

My research employs a mixed methods approach to analyze and design adaptive, human-centered AI systems that mitigate potential harm and maximize benefit for all users. This approach is rooted in stakeholder-engaged research practices throughout the entire lifecycle, from design and instrument development to deployment and evaluation. By combining quantitative and qualitative insights, I aim to create usable, practical AI solutions informed by pluralistic alignment methods that respect the diverse preferences of patients and clinicians while adhering to validated clinical standards.

EHR Data & Clinical Genetics

The electronic health record (EHR) is a rich source of structured and unstructured information for understanding patient risk for disease, disease trajectories, and clinical practice, but the data is oftentimes messy and difficult to extract. I have designed and developed machine learning and clinical natural language processing algorithms to maximize usable information from the EHR to understand and support clinical genetics practice. These include efficient and novel knowledge-integrated approaches that span from rule-based methods to LLM-enabled tools. Projects include:

• Phenotype Detection: Detection of clinically useful phenotypic features from clinical notes of patients suspected of rare Mendelian disorders. [View PDF]
• Workflow Efficiency: Automated extraction of efficiency measures from genetic counseling notes, reducing annotation time by 99%. [Link] [Feature]
• Automated FHx Extraction: Developed family history (FHx) conversational dialog datasets to develop automated FHx extraction methods. Explored knowledge-graph augmented entity linking of extracted FHx to UMLS concepts.

Consumer-Facing Digital Health Chatbots

I am committed to empowering consumers to understand and act on their health by designing and testing consumer-facing digital health tools. By improving health information collection and access through guided, informative chatbots, these tools can lower barriers for diverse populations to engage with their health. Projects include:

• KIT (Family Health History Chatbot): Investigated a novel flow-based chatbot for family history data collection to assess usability, engagement, and report usefulness. Building on this, I've explored preferences for LLM-enabled patient-facing reports, uncovering differential needs that motivate personalized educational material. [Link] | [Watch Video]
• Strolr (Pregnancy Q&A Chatbot): Led a student team to develop a retrieval-augmented-generation chatbot to answer consumer health pregnancy queries, with a design directly informed by semi-structured interviews with key pregnancy and clinical stakeholders.