Research · Publications · Scientific Work

CDEAtlas: Interactive Visualization and Exploration of Common Data Elements at Scale

A first-author poster at the AMIA 2026 Amplify: Informatics Summit presenting CDEAtlas, an interactive visualization system for large-scale exploration of NIH Common Data Elements.

Designing a Visual Analytics Dashboard to Streamline Collaborative Common Data Element Mapping

A first-author poster presented at AMIA 2025, showcasing a clinical NLP dashboard that streamlines concept mapping and improves interpretability for researchers and clinicians.

CDEMapper 2.0: Expediting Collaborative Review and Facilitating Consensus Building During Common Data Elements Mapping

An AMIA 2025 poster presenting CDEMapper 2.0, an extension of CDEMapper that introduces a Delphi-style collaborative review framework with multi-role workflows for conflict resolution and expert consensus formation during NIH Common Data Element mapping.

I designed the user workflow framework illustration that visualizes the collaborative review process, and contributed to the interface design for the system.

CDEMapper: enhancing National Institutes of Health common data element use with large language models

A JAMIA-published study introducing CDEMapper, an LLM-powered tool that combines semantic search, GPT-based recommendations, and human-in-the-loop review to map local data elements to NIH Common Data Elements.

I contributed to evaluation workflows, system usability assessment, and visualization design for the CDE mapping pipeline.

Information Extraction from Clinical Notes: Are We Ready to Switch to Large Language Models?

A JAMIA study evaluating whether LLMs (LLaMA 3, GPT-4o) can replace BERT-based systems for clinical information extraction. Across two institutions, LLMs achieved up to 7% higher F1 on NER and 4% on relation extraction, with improved cross-institutional generalization.

I designed the user-facing interface for Kiwi, the clinical NLP system associated with this study. My work turns the LLM-based information extraction outputs into an inspectable, usable product for biomedical researchers.