DeepSlide enhances presentations with advanced AI delivery support

Ming Yang and a team of researchers have introduced DeepSlide, an innovative AI system designed to enhance the entire presentation process. Unlike traditional slide generators that focus solely on creating visually appealing slides, DeepSlide incorporates a human-in-the-loop, multi-agent approach that optimizes pacing, narrative, and preparation. Key features include a controllable logical-chain planner with specific time budgets, a lightweight content-tree retriever for accurate grounding, and sandboxed execution to ensure high renderability. This comprehensive system addresses both static artifact quality and dynamic delivery excellence through a dual-scoreboard benchmark.

In testing across 20 different domains and audience profiles, DeepSlide consistently matched strong baselines for artifact quality while significantly improving delivery metrics. It enhances narrative flow, pacing precision, and synergy between slides and scripts, offering clearer attention guidance. This allows users to create more engaging and effective presentations, ultimately enhancing scholarly communication and audience engagement. As presentations remain a vital medium for information exchange, tools like DeepSlide represent a critical advancement in optimizing how ideas are communicated.