Reproduce “Building Presentation Intelligence for High-Quality Slide Generation” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Yuhang Xie, Cheng Li, Chenliang Li, Mingyu Xu, Mingwei Sun, Suhang Zheng, Ming Yan, Xiaoxiao Xu and 1 more

The authors present a pipeline for source-grounded presentation generation that combines curated examples and teacher guidance with reinforcement learning from rendered slides. Rule-based rewards assess five visual layout properties. The selected Qwen-3.6-27B checkpoint scores 77.57 on PresentBench and 79.36 on GDPVal PPT, above untrained baselines; for the 35B-A3B model, RL raises aggregate aesthetic scores, though some dimensions decline.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.