PaperExplainAgent

PaperExplainAgent is an interactive reading assistant designed to help researchers and students make sense of dense STEM papers. Given a PDF, the system responds with simple visual or animation-style sketches that could be rendered with tools like Manim. To build this behavior, we build on open-source LLMs with prompting and integration strategies tailored to dense mathematical texts. Our design is motivated by work in cognitive science and math education showing that well-aligned diagrams and dynamic visualizations can reduce cognitive load and improve conceptual understanding, especially for symbol-heavy material. Research suggests that structured, visually grounded responses are preferred to generic textonly explanations and are perceived as clearer and more helpful for understanding unfamiliar results. We conclude by discussing limitations of our current prototype and outlining how tighter integration between interactive reading tools and visualization backends could further support mathematical and STEM researc.