Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Auto-Slides: An Interactive Multi-Agent System for Creating and Customizing Research Presentations

Yuheng Yang, Wenjia Jiang, Yang Wang, Yi Song, Yiwei Wang, Chi Zhang

Abstract

The rapid progress of large language models (LLMs) has opened new opportunities for education. While learners can interact with academic papers through LLM-powered dialogue, limitations still exist: the lack of structured organization and the heavy reliance on text can impede systematic understanding and engagement with complex concepts. To address these challenges, we propose Auto-Slides, an LLM-driven system that converts research papers into pedagogically structured, multimodal slides (e.g., diagrams and tables). Drawing on cognitive science, it creates a presentation-oriented narrative and allows iterative refinement via an interactive editor to better match learners' knowledge level and goals. Auto-Slides further incorporates verification and knowledge retrieval mechanisms to ensure accuracy and contextual completeness. Through extensive user studies, Auto-Slides demonstrates strong learner acceptance, improved structural support for understanding, and expert-validated gains in narrative quality compared with conventional LLM-based reading. Our contributions lie in designing a multi-agent framework for transforming academic papers into pedagogically optimized slides and introducing interactive customization for personalized learning.

Auto-Slides: An Interactive Multi-Agent System for Creating and Customizing Research Presentations

Abstract

The rapid progress of large language models (LLMs) has opened new opportunities for education. While learners can interact with academic papers through LLM-powered dialogue, limitations still exist: the lack of structured organization and the heavy reliance on text can impede systematic understanding and engagement with complex concepts. To address these challenges, we propose Auto-Slides, an LLM-driven system that converts research papers into pedagogically structured, multimodal slides (e.g., diagrams and tables). Drawing on cognitive science, it creates a presentation-oriented narrative and allows iterative refinement via an interactive editor to better match learners' knowledge level and goals. Auto-Slides further incorporates verification and knowledge retrieval mechanisms to ensure accuracy and contextual completeness. Through extensive user studies, Auto-Slides demonstrates strong learner acceptance, improved structural support for understanding, and expert-validated gains in narrative quality compared with conventional LLM-based reading. Our contributions lie in designing a multi-agent framework for transforming academic papers into pedagogically optimized slides and introducing interactive customization for personalized learning.

Paper Structure

This paper contains 15 sections, 3 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Content Understanding and Structuring
  4. High-Fidelity Multi-modal Parsing.
  5. Cognitive-Driven Presentation Planning.
  6. Quality Assurance and Content Adjustment
  7. Multi-modal Generation
  8. Interactive Optimization
  9. Experiments
  10. Evaluation Metrics
  11. User Study 1: Interactive Slide Functions from Learners' Perspective
  12. User Study 2: Comparing Slides and LLM Chat-based Learning from Learners' Perspective
  13. User Study 3: Evaluating Narrative Structure Optimization from Experts' Perspective
  14. Automatic Evaluation: LLM-as-Judge Component Ablation
  15. Conclusion

Figures (3)

  • Figure 1: Auto-Slides' capabilities.Left: Users can generate complete academic presentation slides from an academic paper. Right: Users can iteratively revise the generated slides by providing high-level natural language instructions, enabling efficient and precise slide editing.
  • Figure 2: Overview of Auto-Slides. The multi-agent pipeline transforms papers into slides via three stages: (1) Content Understanding, where Parser and Planner Agents design the slide structure in JSON; (2) Quality Assurance, utilizing Verification and Adjustment Agents to ensure content fidelity; and (3) Generation & Interaction, where Generator and Editor Agents produce LaTeX slides and facilitate human-in-the-loop revisions.
  • Figure 3: Interactive optimization workflow. The Editor Agent processes natural language commands via a ReActyao2023react pipeline, orchestrating content retrieval and LaTeX code modifications to compile updated presentations.