Projects
ActionCue
Visualizing uncertainty-to-action composition for human oversight
ActionCue is a process-transparency visualization that shows how multiple uncertainty conditions are combined into an oversight response. The underlying framework determines whether and how an AI-supported decision may proceed, while keeping that oversight response separate from the substantive domain decision itself.
I designed and built ActionCue to make this composition process explicit and inspectable. The approach was demonstrated through worked cases in healthcare, credit assessment, and disaster forecasting, and compared with confidence-only and data-level uncertainty displays.
UNACORM
Unified evaluation of DNA data-storage codecs
UNACORM provides a common environment for evaluating DNA data-storage codecs across multiple dimensions, enabling systematic and reproducible comparison of their performance and trade-offs.
I co-led the work, designed the application and its user experience, conducted the experiments, evaluated and validated the results, and wrote the manuscript. I also supervised the reimplementation of the codecs used in the evaluation.
DNAsmart
Interactive multi-attribute comparison of DNA data-storage systems
DNAsmart is an interactive visual-analytics tool for comparing DNA data-storage systems across multiple attributes. It allows users to explore how different evaluation criteria and their relative importance affect the ranking of competing approaches.
I conceptualized and implemented the visualization and contributed to writing and reviewing the manuscript.
MetaMP
Unified and auditable membrane-protein annotation
MetaMP brings together membrane-protein annotations from multiple sources in a unified, provenance-aware resource. It supports comparison across databases, exposes disagreement between sources, and provides workflows for benchmark interpretation and quality assessment.
My contribution focused on critical review of the work, interpretation of benchmarking results, and manuscript review and editing.
Guiding Sentiment Analysis with Hierarchical Text Clustering
Exploring public-health discourse through machine learning and visualization
This work combined hierarchical text clustering with sentiment analysis to study large-scale public discourse around face masks during the COVID-19 pandemic. The approach enabled sentiment to be examined within topics and subtopics rather than only at the level of the complete dataset.
I contributed to the analysis of the results and developed visualizations used to interpret the resulting patterns.
CAPT
Context-aware visualization of phylogeny and taxonomy
CAPT links phylogenetic and taxonomic representations to support interactive exploration of relationships between evolutionary structure and biological classification.
I improved the user interface and created the online version of the tool.