projects

A collection of research projects and interactive demos (page currently under construction).

system demonstrations

projects

Projects in this section:

Improving STEM document accessibility

How do we make scientific documents more accessible for everyone? My work places particular emphasis on improving PDF accessibility for blind and low vision users and more recently, on document content accessibility for people with cognitive disabilities and/or IDD. This work spans multiple areas:

  • Scientific PDF Accessibility: We developed SciA11y, a system for converting scientific PDFs into accessible HTML. This project is publicly available as the Paper to HTML web tool, where users can convert scientific PDFs into accessible HTML on demand.
overview of pipeline of the paper to html conversion system, showing how a user can upload a PDF, the system runs various models to extract text, structure, images, and tables, and reconstruct a tagged html
  • Alt Text Generation and Analysis: We introduced FigurA11y, a tool that integrates an AI assistant to help authors write scientific alt text. We’ve also released datasets of alt texts from HCI publications and developed automated methods for analyzing alt text quality to understand current practices and gaps.

  • Accessibility Research Meta-Analysis: Through surveying and analyzing trends in the literature, we better understand accessibility research practices and citation patterns. This work includes bibliometric analyses of citation diversity in accessibility and HCI research, and a survey examining how research communities reference and build upon accessibility work. Our current investigation reveals the declining accessibility of scholarly PDFs over the last decade, and how publishing model and platform changes affect document accessibility.

Relevant Publications:

Blog Posts: SciA11y part I and part II

Ongoing Work:

  • Developing more robust AI-assisted accessibility tools
  • Measuring and addressing the ongoing accessibility problem in scholarly publishing
  • Creating guidelines and standards for accessible scientific communication
  • Going beyond document accessibility to address issues of content accessibility
↑ back to top

AI-assisted Literature Review

How can we use AI to help researchers discover, analyze, and synthesize scientific literature more effectively? My work focuses on developing tools and methods that assist researchers in navigating the ever-growing body of scientific literature, with particular emphasis on biomedical research and scientific claim verification. This work spans multiple areas:

  • Literature Discovery and Synthesis: Through projects like CHIME and MS², we’re developing tools to help researchers organize and synthesize scientific literature. CHIME provides LLM-assisted hierarchical organization of studies for literature reviews, while MS² focuses on multi-document summarization of medical studies. I have also contributed heavily to the development of the Semantic Scholar Open Research Corpus (S2ORC) and many other open scholarly datasets.

  • Scientific Claim Verification: We developed several systems for verifying scientific claims against literature. These frameworks help identify supporting or refuting evidence for scientific claims, detect contradictory statements, and assess claim consistency across papers. Our work includes methods for zero-shot claim verification and automated claim generation from scientific texts.

Relevant Publications:

Ongoing Work:

  • Developing better evaluation metrics for scientific long-form generation
  • Creating tools to help researchers organize and synthesize literature
  • Investigating the role of AI in scholarly information access
↑ back to top