CRui5in/paper-ppt-agent

What it solves

It automates the creation of academic presentation slides from research papers. Instead of manually summarizing a PDF or TeX file and designing layout, the tool handles content extraction, structural planning, and visual design to produce an editable PowerPoint file.

How it works

The project uses a multi-agent pipeline consisting of a Strategist, an Executor, and a Critic. The Strategist plans the structure, the Executor generates the content and layout, and the the Critic performs quality assurance (QA) to detect issues like text overflow or low contrast, triggering fixes if necessary. It supports various LLMs (OpenAI, Anthropic, Gemini, DeepSeek) and can integrate with local agents like Claude Code or Codex for generation and template analysis.

Who it’s for

Researchers, students, and academics who need to quickly transform their research papers into professional presentation decks.

Highlights

• Multi-Agent Workflow: A three-stage pipeline (Strategist $\rightarrow$ Executor $\rightarrow$ Critic) for integrated content and layout generation.
• Visual QA: Uses multimodal LLMs to render slides as images to check for layout errors and contrast issues.
• Deep Research: Enhances content with external data from arXiv, Semantic Scholar, and the web.
• Interactive Editing: Includes a built-in visual PPT editor (based on PPTist) for manual adjustments to pages, fonts, and notes.
• Template Support: Allows importing existing PPTX files as templates or using AI to analyze and template them.
• Version Control: Automatically archives snapshots of each iteration for comparison and rollback.