Sanghani Center Student Spotlight: Xuxin Tang

Graphic is from the paper “Steering LLM Summarization with Visual Workspaces for Sensemaking”

In her Ph.D. research in computer science, Xuxin Tang is focused on developing interactive systems where humans and large language models (LLMs) — often seen as “two black boxes” — can work together more effectively to support analytical tasks and writing, and enhance sensemaking. 

A real-world example of this is her work on designing visual workspaces that help users interact with LLMs for summarizing and analyzing large information sets. 

“Imagine a journalist researching a complex topic with hundreds of sources,” said Tang.  By using a visual workspace, they can organize, analyze, and explore information with the help of an LLM, which generates summaries, finds connections, and adapts based on the user’s evolving needs. This approach not only makes the analysis more efficient but also helps users retain oversight and fine-tune the AI’s contributions, ensuring that the final output aligns with their insights and goals.”

Tang, who earned a bachelor’s degree in engineering and a master of science from Wuhan University, China, is advised by Chris North at the Sanghani Center.

“The collaborative and interdisciplinary research environment at the center encourages innovation and creativity,” said Tang. “By bringing together students and faculty from diverse fields, being at the Sanghani center allows me to gain insights beyond my primary focus areas and approach my research in human-LLM collaboration and visual analytics from fresh perspectives.”

In October, Tang published and presented the paper, “Steering LLM Summarization with Visual Workspaces for Sensemaking” at the IEEE VIS 2024 workshop — NLVIZ: Exploring Research Opportunities for Natural Language, Text, and Data Visualization.

Tang’s attraction to his area of research grew from her background in machine learning and recommendation systems, where he saw both the incredible potential and the challenges of AI, she said. “I realized that while these models can process massive amounts of information and generate insights, they often operate as ‘black boxes,” limiting human oversight and understanding. This sparked my interest in finding ways to make AI more interpretable and collaborative.”

Projected to graduate in 2027, Tang said that she will pursue positions in both academia and industry to further advance her research.