MemSlides: A Hierarchical Memory Driven Agent Framework for Personalized Slide Generation with Multi-turn Local Revision
Ye Jin, Yangyang Xu, Jun Zhu, Yibo Yang
arXiv:2606.17162 · 176▲ · cs.CL, cs.HC, cs.MA
View on arXiv →Premium readers get an interactive explainer for this paper — a figure you can poke at, not just read.
What it is
MemSlides is an agent framework for generating and editing presentation slides that splits memory into three parts: long-term user profile memory (persistent style/layout preferences per user and intent), tool memory (reusable execution traces for editing tools), and session-scoped working memory (temporary constraints across revision turns). It pairs this with a Plan-Act-Guard revision pipeline that scopes edits to the smallest affected slide region rather than regenerating the whole deck on each feedback turn.
Why it matters
For anyone building document or slide generation agents, the practical lesson is that treating personalization as one undifferentiated chat buffer is fragile; separating persistent preferences from session state and from tool execution history reduces context pressure and stops small edits from overwriting already-aligned content. The scoped local-edit approach also cuts wasted tool time, which matters for latency and API cost in multi-turn editing loops.
Practical takeaway
If you build multi-turn editing agents, consider a localized patch strategy (edit the targeted region with verified coverage checks) instead of full regeneration, plus a memory tier that caches successful tool-call chains to retrieve before similar future calls. Watch for this hierarchical memory pattern showing up in slide and document tooling.
Key result
On a diagnostic matched-pair modify evaluation (only 9 pairs, controlled proxy tasks, not real users), tool-memory injection cut geometric-mean core tool time to 0.327x of the no-injection baseline and reduced time to first correct edit from 609.5s to 242.5s; the strongest paired evidence was Strict Verify (0.310 to 0.534) and tool time ratio, both with sign-test p=0.0195, while completion and latency gains were only directional.
Subscribe
Get the next issue.
Free. One email a week. Unsubscribe any time: no account, no dark patterns.