How Multi‑Agent Architectures Power the Next Generation of Recommendation and Search Systems

Based on a technical sharing by Xing Shaomin, the head of Alibaba Cloud AI Search, the article first outlines the challenges of high‑concurrency, multi‑modal data, and complex multi‑hop queries, and presents the Agentic RAG architecture that evolves from a single agent to a multi‑agent system. The design coordinates planning, retrieval, and generation modules, and adopts a mixed‑recall strategy that combines vector, text, database, and graph sources to improve coverage and accuracy.

The discussion then shifts to the evolution of recommendation systems from deep‑learning models to large language model (LLM)‑driven AI agents, using Huawei Noah’s KAR project as a case study. It describes how factorized prompting and a multi‑expert knowledge adapter map semantic knowledge into the recommendation embedding space, and how the multi‑expert network balances text feature dimensionality with real‑time constraints. Reported results include a 1.5 % AUC increase and online A/B‑test data.

Next, the article details Baidu’s generative ranking model GRAB, which replaces traditional discrete feature engineering with an end‑to‑end sequence generation approach inspired by LLM scaling laws and the Transformer architecture. A Q‑Aware RAB causal attention mechanism introduces query‑aware relative bias for adaptive modeling of complex interactions and temporal signals. The paper also explains the STS two‑stage training pipeline, heterogeneous token representations, dual‑loss stacking, and KV‑Cache optimizations that enable high‑throughput online inference, and it provides quantitative business benefits observed after full deployment.

Finally, the piece lists the eight chapters of the ebook “Intelligent Agent Architecture and Practice: Building the Next‑Generation Recommendation and Search Systems”, covering topics such as multi‑agent interaction for AI for good, knowledge discovery with LLM agents, observability of OpenAI Swarm‑style systems, and practical guides for Elasticsearch‑based vector search and RAG applications.