Download
PPTAgent is a research system for generating and evaluating slide decks that goes beyond simple text-to-slides. It follows a two-stage, edit-based workflow: first it analyzes reference presentations to infer slide roles and structure, then it drafts an outline and iteratively performs editing actions to produce new slides. The project includes both the generation agent and an evaluation framework, PPTEval, to score content quality, design, and coherence. The repository highlights the EMNLP 2025 paper and provides links to resources for replication and study. The approach reflects human presentation practice—plan, draft, then refine with edits—yielding more coherent decks than direct one-shot generation. Community interest and stars suggest strong uptake for research and tooling around presentation automation.
Features
- Two-stage pipeline that plans and then edits slides
- Slide-type and content-schema inference from references
- Iterative editing actions for structure and design quality
- PPTEval framework to assess content, design, and coherence
- Reproducible code and resources aligned with the paper
- Agentic workflow suitable for integration with MCP clients
Project Samples
