Summary CRui5in/paper-ppt-agent automates the creation of editable PowerPoint presentations from academic papers (PDF or LaTeX). It targets researchers and professionals needing to quickly generate slides from dense documents, offering AI-driven content extraction, structure planning, layout design, and visual quality checks.

How It Works A multi-agent pipeline (Strategist → Executor → Critic) processes paper inputs (PDF or TeX, compressed TeX recommended) to distill content, plan structure, and generate layouts. The Critic performs static analysis for layout issues (overflow, overlap) and experimental multimodal QA on rendered images. An integrated icon library supports RAG-based semantic search (Gemini Embedding) for contextually relevant icon insertion. The system supports multi-language output and configurable presentation details.

Quick Start & Requirements Simplified setup uses .\start-dev.bat (Windows) or sh start-dev.sh (Linux) for automated dependency installation and service startup. Manual setup requires Python 3.11+ and Node.js 18+ with npm. Crucially, an API key from a supported model provider (OpenAI, DeepSeek, Anthropic, Gemini) is mandatory; custom BaseURLs are compatible. A Gemini API key is optional for icon RAG. Local access is via http://127.0.0.1:5173 (frontend) and http://127.0.0.1:8000 (backend).

Highlighted Details

• Editable PPT Generation: Creates editable PowerPoint decks from academic papers.
• Multi-Agent Workflow: Strategist → Executor → Critic pipeline for content and layout processing.
• Visual Quality Assurance: Static analysis (layout, contrast) and experimental multimodal QA, with SVG snapshot comparisons.
• Intelligent Icon Integration: Icon library with RAG-based semantic search.
• Iterative Refinement: Supports feedback for structure adjustments with version history.
• Comprehensive Observability: Real-time logs, token usage tracking, detailed Critic reporting.

Maintenance & Community This is an open-source tool. Support is via GitHub Issues or email (qinruoxuan2018@gmail.com). No specific community channels or maintainer details are provided.

Licensing & Compatibility Released under the permissive MIT License, allowing broad use, including commercial and closed-source integration.

Limitations & Caveats Positioned as an academic research assistant; AI-generated content is for reference only. Users are responsible for output accuracy and compliance. Visual QA is experimental.