Automatic conversion of academic papers into slide decks is inherently iterative, requiring cycles of generation, critique, and revision. Existing multi‑agent systems introduce internal critique‑revise loops but remain largely opaque to users, or they accept feedback only after a complete deck is produced, which hampers user participation in narrative flow, content allocation, and emphasis decisions.\ \ ConvDeck addresses this gap with a multi‑agent pipeline that distributes interaction across distinct stages. Conversational feedback loops are embedded at outline creation, content filling, and visual layout stages, allowing users to intervene precisely where each type of decision is made. Agents equipped with think, speak, and act capabilities either apply edits directly or engage in dialogue to clarify feedback and explore revision options.\ \ Evaluation shows that stage‑specific conversational feedback improves user‑goal satisfaction while preserving narrative coherence, content quality, and visual presentation.\ \ Review