Researchers propose MemSlides, a hierarchical memory framework for personalized slide generation agents

MemSlides is a hierarchical memory framework for personalized presentation agents that separates long-term memory from working memory, further dividing long-term memory into user profile memory and tool memory. User profile memory stores intent-conditioned profiles for initial personalization, while working memory carries active preferences and session constraints across revision rounds. Tool memory stores reusable execution experience to support reliable localized editing.

The framework pairs this memory design with scoped slide-local revision, enabling targeted updates to act on the smallest affected region instead of regenerating the full deck. This approach aims to improve efficiency and precision in multi-turn revision workflows.

In controlled experiments, user profile memory improved persona-alignment judgments on a multi-persona, multi-intent profile bank. Tool-memory injection improved closed-loop modify behavior in diagnostic matched-pair settings, and qualitative cases illustrated working memory’s ability to carry over preferences across sessions.

The authors argue that effective personalization in presentation authoring depends on separating persistent user profiles, session-level working memory, and reusable execution experience across both generation and localized revision phases.