Designing Next‑Gen Recommendation and Search with Intelligent Agent Architecture

This piece introduces an ebook that gathers eight technical chapters on intelligent‑agent architectures and their application to modern recommendation and search systems.

Alibaba Cloud AI Search – Agentic RAG

The author summarizes a talk by Alibaba Cloud AI Search lead Xing Shaomin, describing challenges such as high concurrency, multimodal data, and multi‑hop queries. The solution evolves from a single‑agent to a multi‑agent framework that coordinates planning, retrieval, and generation modules to understand complex intents. A multi‑path retrieval layer mixes vector, text, database, and graph recall to improve coverage and accuracy, and GPU‑accelerated indexing and query quantization are compared, showing measurable speed‑up.

Huawei Noah Recommendation – LLM Integration

The chapter reviews the transition from deep‑learning recommenders to large‑language‑model (LLM) and AI‑Agent eras, highlighting problems of noisy implicit feedback, limited semantic understanding, and intent mining. It contrasts list‑based and conversational recommendation flows, and details the KAR project where factorized prompting and a multi‑expert knowledge adapter map semantic knowledge into the recommendation embedding space. The design balances text feature dimensionality with real‑time constraints, and an online A/B test reports a 1.5 % AUC lift.

Baidu GRAB – Generative Ranking for Ads

The Baidu commercial tech team’s GRAB model replaces traditional DLRM pipelines by end‑to‑end generative sequence modeling of user behavior and target ads, leveraging LLM scaling laws and Transformer architecture. A Q‑Aware RAB causal attention mechanism introduces query‑aware bias for adaptive modeling of complex interactions and temporal signals. The paper also explains the STS two‑stage training algorithm, heterogeneous token representations, a dual‑loss stacking strategy, and KV‑Cache optimizations for high‑concurrency inference, together with quantified business gains after full deployment.

Elasticsearch Vector Search and RAG

One chapter demonstrates how to use Elasticsearch for vector search and to build Retrieval‑Augmented Generation (RAG) applications, detailing index construction, query pipelines, and integration points.

Table of Contents

1. Multi‑agent interaction systems for AI‑for‑Good<br/>2. Knowledge discovery and data‑science with LLM agents<br/>3. Observability of OpenAI Swarm at SF Tech<br/>4. Huawei Noah: recommendation evolution and LLM practice<br/>5. GRAB: Baidu’s generative ad ranking model<br/>6. Elasticsearch vector search and RAG<br/>7. Alibaba Cloud AI Search Agentic RAG practice<br/>8. From big data to big models: frontier of search‑recommendation