Understanding people.
Designing for better care.

Understanding PPE use and alert design

Design explorations for just-in-time PPE compliance alerts in interdisciplinary medical settings.

Video-based PPE compliance monitoring

Design and early evaluation of a video-based monitoring application for time-critical care.

LLM-Powered Medical Chatbot with RAG

View project on GitHub ↗

Exploring how prompt engineering and retrieval-augmented generation affect medical chatbot responses, through technical and human-centered evaluation.

I built and evaluated three configurations to distinguish the effects of prompt engineering from retrieval. The project asks whether adding relevant doctor-response examples improves an LLM’s answers—and whether technical scores capture qualities that matter under time pressure.

Python · GPT-4o-mini · Sentence Transformers · FAISS · Gradio

Live demo currently offline.

Read project overview

What I built

A Python medical question-answering prototype using GPT-4o-mini, sentence-transformer embeddings, and FAISS retrieval over 50,000 doctor responses from the HealthCareMagic dataset.

How I evaluated it

Compared an original-prompt baseline, an improved prompt without retrieval, and an improved prompt with RAG on the same 20 questions. Evaluation combined LLM-as-judge scores, ROUGE, and a manual HCI rubric.

What I learned

The improved prompt scored highest in the LLM-judge evaluation; RAG scored highest on the manual HCI rubric. Conciseness remained a weakness across configurations. Retrieval did not consistently outperform prompt engineering alone in this small comparison.

Exploratory results

Across 20 questions, LLM-judge totals were 11.15/12 for the baseline, 11.60/12 for the improved prompt, and 11.55/12 for RAG. On the manual HCI rubric applied to 10 responses per configuration, totals were 10.1/15, 11.7/15, and 12.3/15 respectively.

These scores reflect this small project evaluation, not demonstrated clinical effectiveness. The different rankings show why a single automated score is insufficient for assessing interface and response quality.

How the prototype works

Question → sentence-transformer embedding → FAISS search for three similar doctor responses → retrieved context plus question sent to GPT-4o-mini → response displayed in a Gradio chat interface.

Design implications & next steps

Constrain response length, broaden evaluation, and test with clinicians before drawing conclusions about clinical suitability. A stable hosted demo remains to be configured.

Project scope

Independent project. Small evaluation set, one non-clinical HCI evaluator, and no clinician testing or clinical validation.

Latent tuberculosis treatment completion

Master’s project comparing completion of latent tuberculosis therapy among HIV-coinfected individuals and other high-risk groups.

COVID-19 outbreaks in nursing homes

Retrospective public health surveillance analysis and discussion of nursing home outbreaks.

Cervical cancer risk prediction

Exploring cervical cancer risk classification using XGBoost.

Early diabetes prediction

Comparing logistic regression, decision tree, and random forest classifiers.

Cardiovascular disease & cancer-cell classification

Automated machine learning classification using AWS AutoGluon.

Medical insurance premium analysis

A regression project using artificial neural networks.

Global COVID-19 forecasting

Time-series forecasting of global COVID-19 cases using Meta Prophet.

Canadian immigration visualization

Exploring immigration to Canada from 1980 to 2013.

Airline performance dashboard

A Python dashboard for annual airline performance data.

King County house-price prediction

Housing-price data analysis with Python.

Historical stock & revenue analysis

An analysis project focused on historical stock and revenue data.

Data science capstone

A data science capstone notebook project.

Portfolio website development

Zootopia

An app concept supporting zoo visits with exhibit information, maps, ticketing, and accessibility information.

Talk & Cook — Voice-Based Cooking Interface

A voice-interface prototype for preparing Nigerian Jollof rice, with step-by-step guidance and hands-free commands.

Smart-Ordering Table

An interactive dining-table concept for ordering, requesting assistance, and payment.

Bounding Boxes in Practice: Design and Early Evaluation of a Video-Based PPE Compliance Monitoring Application for Time-Critical Care ↗

Christine D. Ikponmwonba, Vidhi Shah, Sifan Yuan, Ivan Marsic, Randall Burd & Aleksandra Sarcevic

Understanding Personal Protective Equipment Use in Interdisciplinary Medical Settings: Design Explorations for Just-in-Time Compliance Alerts ↗

Aleksandra Sarcevic, Eleanor Wood, Katherine Ann Zellner, Christine D. Ikponmwonba, Mary Suhyun Kim, Ivan Marsic & Randall S. Burd

Intermittent hypoxia and caffeine in infants born preterm: the ICAF Randomized Clinical Trial ↗

Adverse effects of COVID-19 pandemic on a multicenter randomized controlled trial ↗

SUID Tissue Consortium: Advancing Medical Research through Postmortem Tissue Donation ↗

C. Ikponmwonba & B. McEntire · American SIDS Institute

Developing team skills competencies for the global health context: interprofessional predeparture training for health science students engaging in global health projects at the University of Florida ↗

E. L. DeVos, C. Ikponmwonba, K. Patel, S. L. McKune, J. W. Grigg & A. Blue

How the interdisciplinary leads to the innovative in global health: learning from the Global Health Case Competition at the University of Florida ↗

S. L. McKune, C. Ikponmwonba & M. Alabduljabbar

Poster presentation · 10th Annual Drexel Emerging Graduate Scholars (DEGS) Research Showcase ↗

Adapting Hand Hygiene Monitoring Technologies to PPE Compliance: An Initial Exploratory Literature Review

SUID Tissue Consortium: Advancing Medical Research through Postmortem Tissue Donation

SUID Tissue Consortium: Introduction

Session Moderator — ISPID: International Conference on Stillbirth, SIDS, and Baby Survival

Conference program ↗

Sudden Unexpected Infant Deaths in Nigeria — Panel: SUID in Different Countries: Rates and Prevention Strategies

Global Health Case Competition at the University of Florida

Peer Reviewer

CHI ’27: ACM Conference on Human Factors in Computing Systems.
CHI ’26: ACM Conference on Human Factors in Computing Systems.

Graduate Assistant, Interactive Systems for Healthcare (IS4H) Research Lab ↗

Healthcare HCI research focused on clinical workflows and the design and evaluation of AI-based PPE alerts.

Graduate Assistant, Global Health Council

Mentor: Sarah McKune, PhD ↗

Supported council meetings and annual global health symposia; contributed to reports, posters, and abstracts; and represented the council through poster presentations at local and national conferences.

Related poster presentation
McKune, S. L., Ikponmwonba, C., & Alabduljabbar, M. How the interdisciplinary leads to the innovative in global health: learning from the Global Health Case Competition at the University of Florida ↗
Seventh Annual Consortium of Universities Global Health Conference · April 8, 2016 · San Francisco, CA

Teaching Assistant

College of Public Health & Health Professions

• HSC 3502 — Survey of Diseases and Disabilities I
• OTH 3201 — Applied Human Development II

Research focus: Designing PPE alerts.

• Abrian Jackson-Wright

Research focus: Redesigning the PPE user interface.

• Liya Giby

Research focus: Qualitative analysis of PPE alert design interviews.

• Arya Patel
• Sara Karabeg

Research focus: Prototyping the PPE alert interface, with co-authorship of Bounding Boxes in Practice ↗.

• Vidhi Shah

American SIDS Institute (ASI) ↗

Projects Director — Analytics & Operations

Led multisite healthcare research operations, bringing together hospitals, medical examiner offices, donor services, and research partners.

• Led case-reporting coordination and workflow analysis for the SUID Tissue Consortium across 8 medical examiner districts.
• Supported collaboration among medical examiner offices, donor-service partners, and the tissue bank, and led training to support partner participation.
• Developed operating procedures, case-tracking workflows, data-quality checks, and research training materials.
• Reviewed and reconciled clinical, autopsy, toxicology, consent, and biospecimen records to support data completeness and analysis.
• Coordinated device deployment and data collection across 14 clinical sites for the Intermittent-hypoxia Caffeine (ICAF) study.

  Related publications

  2025 · Archives of Disease in Childhood — Fetal and Neonatal Edition
  Intermittent hypoxia and caffeine in infants born preterm: the ICAF Randomized Clinical Trial ↗

  2023 · Journal of Perinatology
  Adverse effects of COVID-19 pandemic on a multicenter randomized controlled trial ↗

• Facilitated cross-functional coordination and training to support consistent research operations.

Questions about the SUID Tissue Consortium or access to tissue for research?
Medical examiner and donor-service partners, as well as researchers, can visit the American SIDS Institute website ↗ for contact information.

American Association of SIDS Prevention Physicians (AASPP) ↗

International Society for the Study and Prevention of Perinatal and Infant Death (ISPID) ↗

Planning Committee Member · 2018–2023

Scientific Committee · Planning the 2023 conference in Italy

Member · 2017–present

Epidemiology Panel Moderator · September 2021 · Virtual conference

Session Moderator · June 2018 · Glasgow, Scotland, United Kingdom
International Conference on Stillbirth, SIDS, and Baby Survival

Conference program ↗