The Harvard Data Science Initiative is pleased to announce support for six new faculty research projects:
Accelerating Improvement in Medical Education Delivery (AI-MED)
Kevin Croke, Associate Professor of Global Health, proposes the first field deployment of a self‑administered, LLM‑driven clinical‑vignette platform to measure primary care provider competencies and evaluate education interventions in Vietnam. Addressing the high cost and limited scalability of traditional, in‑person vignettes, the project uses a conversational survey interface that role‑plays standardized “patients,” localizes on the fly to Vietnamese, and supports baseline and one‑year follow‑up assessments for the 2025 StITCH cohort in Phú Thọ province.
A follow on from a summer 2025 pilot and validation conducted through the HMS Global Primary Health Care program, the team will integrate OpenAI‑based chatbots into an existing survey platform to produce complete transcripts for human or automated grading at negligible marginal cost. Core cases adapt Service Delivery Indicators vignettes to the Vietnamese context, creating scalable, repeatable measures that can inform medical education and primary‑care reforms.
Learn more about Professor Croke’s research.
UK Biobank Data Access and DNAnexus Computing Infrastructure for Multi-Omics Neurodegenerative Disease Research
Peng Gao, Assistant Professor of Environmental Health and Exposomics, and Chirag Patel, Associate Professor of Biomedical Informatics, are building on completed analyses that uncovered 86 significant environmental exposures and key proteomic biomarkers for dementia. The team will validate findings and develop clinically actionable risk‑models using newly released proteomics for 50,000 participants and enhanced neuroimaging across the UK Biobank’s 500,000‑person cohort.
Learn more about Professor Gao’s research and Professor Patel’s research.
Attention versus Sentiment: What Drives Household Investment Performance?
Daniel Graves, Assistant Professor of Economics, is developing a structural, random‑coefficients model of household stock demand that separates three forces; limited attention (which stocks enter investors’ consideration sets), stock‑specific sentiment (idiosyncratic expected returns), and beliefs about observable characteristics (value, momentum, beta), to test whether retail investors are collectively “skilled.”
Learn more about Professor Graves’s research.
Clinician-in-the-Loop Explainability Evaluation Tool for Medical Imaging AI Models
Ryan Harari, Instructor in Radiology, is developing a lightweight platform to systematically test which AI explanation formats clinicians actually find most useful and trustworthy when interpreting imaging‑based predictions. Addressing a key barrier to real‑world adoption, the project focuses on clinician trust and decision‑support rather than model accuracy. Building on an IRB‑approved knee osteoarthritis MRI study, the interface will present de‑identified images, model outputs, and multiple explanation types side‑by‑side; clinicians will choose a preferred explanation, rate confidence, and optionally give feedback, with all interactions logged securely and anonymously.
Learn more about Doctor Harari’s research.
Incidental Representation: How Special Interests Advance a Broader Public
Gautam Nair, Assistant Professor of Public Policy at Harvard Kennedy School, is refining AI methods to detect “incidental representation”–moments when organized, concentrated interests (e.g., firms, industry groups) unintentionally advance the interests of broader, diffuse publics such as consumers, low‑income citizens, or the environment. The team will apply modern NLP and generative‑AI techniques to the ProQuest TDM Studio archive of U.S. Congressional hearings (1900–present) to map when and how special‑interest witnesses explicitly or implicitly speak for the wider public across trade, climate, social, and antitrust policy domains.
Learn more about Professor Nair’s research.
Identifying Trust-Building Characteristics from A Multi-Modal Social Media News Dataset
Amanda Yarnell, Dependent Lecturer on Social & Behavioral Sciences and Senior Director of the Center for Health Communication, seeks to pinpoint which visual and creative choices in social‑video news (especially TikTok) foster audience trust as attention shifts from legacy institutions to individual creators. Filling a gap in prior work that often uses single‑item trust measures, the project employs a validated, multidimensional framework, comparing independent “news creators” with professional newsrooms. The study combines automated analysis of objective video properties with subjective crowd ratings to link design choices to perceived trustworthiness.
Learn more about Amanda’s research.
All projects are supported through the HDSI Faculty Special Projects Fund.