MemSlides: A Hierarchical Memory Driven Agent Framework for Personalized Slide Generation with Multi-turn Local Revision

2026-06-15 · Source: Computation and Language · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Human-Computer Interaction · Depth: Expert, quick

Summary

MemSlides is a hierarchical memory-driven agent framework designed for personalized slide generation with multi-turn local revision. It separates long-term memory into user profile memory and tool memory, distinct from working memory. User profile memory stores intent-conditioned profiles for initial personalization, while working memory retains active preferences and session constraints across revision rounds. Tool memory stores reusable execution experience for reliable localized editing. MemSlides integrates this memory design with scoped slide-local revision, enabling targeted updates on the smallest affected regions instead of full deck regeneration. Experiments show user profile memory improves persona-alignment, tool-memory injection enhances closed-loop modify behavior, and working memory effectively carries over preferences, indicating that effective personalization in presentation authoring relies on separating these memory types.

Key takeaway

For AI Engineers developing personalized content generation agents, adopting a hierarchical memory framework like MemSlides is crucial. Your systems should separate persistent user profiles, session-level working memory, and reusable execution experience to improve persona-alignment and reliably handle multi-turn revisions. This approach ensures preferences are maintained and edits are localized, leading to more effective and user-friendly generative AI applications.

Key insights

MemSlides employs hierarchical memory and local revision for personalized, agent-driven slide generation.

Principles

• Separate long-term and working memory.
• Divide long-term memory into user profile and tool.
• Apply scoped slide-local revision.

Method

MemSlides combines a hierarchical memory system (user profile, working, tool memory) with scoped slide-local revision to manage user preferences and execution experience, enabling personalized, multi-turn presentation generation and targeted edits.

In practice

• Store intent-conditioned user profiles.
• Retain session constraints in working memory.
• Isolate edits to smallest affected regions.

Topics

• Hierarchical Memory
• Agent Frameworks
• Presentation Generation
• Personalized AI
• Multi-turn Interaction
• Human-Computer Interaction

Best for: Research Scientist, AI Scientist, NLP Engineer, AI Engineer

Related on AIssential

• Memp: Exploring Agent Procedural Memory — Procedural memory, dynamically built and updated, significantly enhances LLM agent performance and efficiency across diverse tasks.
• Using MemAlign to Improve Evaluation of Traditional Machine Learning in Genie Code — MemAlign significantly improves LLM judge alignment with human experts using dual-memory from natural language feedback.
• AI Agents of the Week: Papers You Should Know About — Advancements in AI agents focus on improving memory, planning, collaboration, and reliability through novel architectures and practical tools.
• MemCollab: Cross-Agent Memory Collaboration via Contrastive Trajectory Distillation — MemCollab enables LLM agents to share memory by distilling agent-agnostic reasoning constraints from diverse reasoning trajectories.
• MemRefine: LLM-Guided Compression for Long-Term Agent Memory — LLM-guided memory compression, prioritizing factual content over surface similarity, effectively manages agent memory within fixed budgets.
• Agentic Recommender System with Hierarchical Belief-State Memory — Hierarchical memory and adaptive lifecycle management significantly enhance agentic recommender system performance.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computation and Language.