Source record

Published by arXiv cs.MA (T1 source). The original is at https://arxiv.org/abs/2610.11375.

Pipeline notes

The summary and note below are generated by the signal pipeline — they are Beyond Desk’s reading, not quotations from the source.

Proposes a multi-agent, multimodal RAG framework for personalized conversational shopping that decomposes dialogue state tracking, retrieval, preference-aware reasoning, and response generation. On an Amazon Reviews 2023 benchmark, retrieval variants outperform a no-RAG baseline (4.82 vs 3.74), and a small n=5 user study shows higher perceived personalization (4.60 vs 2.20).

Provides a concrete role-decomposition blueprint with released code and trajectory-level evaluation protocol, serving as a reference architecture for multi-agent RAG systems beyond the shopping domain.

Cited by

No citations on record.