Agent Architecture in Action: Building Next‑Gen Recommendation and Search Systems

Agentic RAG in Alibaba Cloud AI Search

The piece summarizes a technical share by Xing Shaomin of Alibaba Cloud AI Search, outlining challenges such as high concurrency, multimodal data, and multi‑hop queries. It describes the evolution from a single Agent to a multi‑Agent system that coordinates planning, retrieval, and generation modules to precisely understand complex intents. The multi‑path retrieval chain mixes vector, text, database, and graph recall to boost coverage and accuracy. Optimization details include GPU‑accelerated indexing and query quantization, with reported speed‑up figures, and extensions like NL2SQL and multimodal search. Full architecture diagrams and performance evaluation data are referenced.

Huawei Noah's Recommendation System Evolution

This section reviews the transition from deep‑learning recommenders to large‑language‑model (LLM) and AI‑Agent eras. It identifies core challenges—noisy implicit feedback, weak semantic understanding, and difficulty mining user intent. Using the KAR project from Huawei Noah’s Ark Lab as a case study, the article explains how factorized prompting and a multi‑expert knowledge adapter map semantic knowledge into the recommendation embedding space. The multi‑expert network balances text feature dimensionality with real‑time constraints. Experiments show a 1.5% AUC lift and online A/B‑test results, with detailed model diagrams provided.

Baidu GRAB: Generative Ranking for Ads

The Baidu commercial tech team’s GRAB model replaces traditional DLRM pipelines with an end‑to‑end generative sequence model that jointly encodes user behavior and target ads in a unified representation space, inspired by LLM scaling laws and Transformer architecture. A novel Q‑Aware RAB causal attention mechanism introduces query‑aware relative bias for adaptive modeling of complex interactions and temporal signals. The paper details a two‑stage STS training algorithm to improve efficiency and mitigate over‑fitting, heterogeneous token representations for hot‑start, a dual‑loss stacking strategy, and KV‑Cache optimizations for high‑concurrency inference. Quantitative business gains after full deployment are reported.

Elasticsearch Vector Search and RAG Applications

The final part outlines how Elasticsearch can be leveraged for vector search and to build Retrieval‑Augmented Generation (RAG) applications, describing the indexing pipeline, query workflow, and integration points for large‑scale semantic retrieval.