LitPivot: Developing research ideas through dynamic contextualization within the literature landscape

Researchers often adjust their ideas after encountering prior work, a process called a literature-initiated pivot. LitPivot, developed with the University of Pennsylvania, streamlines this by organizing literature around specific parts of a proposal, such as the problem, contribution, or evaluation plan. The system retrieves papers via Asta Paper Finder and provides critiques with supporting excerpts, allowing researchers to revise their drafts with targeted feedback.

In a study with 17 researchers, domain experts rated proposals developed using LitPivot as better grounded in the literature compared to those created with a text editor and an AI chat tool. Participants also reported greater novelty and usefulness in their revised ideas, alongside a stronger understanding of related work. The tool segments proposals into facets and groups papers by relevance to each segment, enabling precise assessments and iterative refinements.

LitPivot’s workflow involves selecting a facet of a proposal, reviewing grouped papers, and requesting an assessment that pairs critiques with suggested revisions. The system flags outdated sections, such as an evaluation plan no longer aligned with a revised contribution, and allows researchers to accept, edit, or reject suggestions. This process helps researchers refine ideas by continuously realigning them with the most relevant literature.

The tool will be presented at UIST 2026 in Detroit. It builds on prior AI research tools like Scideator and IdeaSynth, which support idea generation and refinement, by integrating literature selection directly into the revision process. LitPivot aims to help researchers turn early ideas into well-supported contributions by making the interplay between evidence and revision explicit.