Designing Next‑Gen Recommendation and Search Systems with Agentic Architectures

The piece summarizes Alibaba Cloud AI Search’s Agentic Retrieval‑Augmented Generation (RAG) practice, which addresses high‑concurrency, multimodal, and multi‑hop query scenarios. It describes the evolution from a single‑agent to a multi‑agent system, where planning, retrieval, and generation modules cooperate to interpret complex intents. The multi‑path retrieval layer mixes vector, text, database, and graph recalls to boost coverage and accuracy, and the authors report GPU‑accelerated indexing and query quantization that yields measurable speed‑up.

It then reviews the technical evolution of recommendation systems from deep‑learning models to large language model (LLM) and AI‑Agent eras. Core challenges such as noisy implicit feedback, limited semantic understanding, and difficulty mining user intent are examined. The authors compare list‑based and conversational recommendation paradigms, and detail how LLMs serve as feature enhancers and pipeline components. Using Huawei Noah’s KAR project as a case study, the article explains factorized prompting and a multi‑expert knowledge adapter that maps semantic knowledge into the recommendation embedding space while balancing text‑feature dimensionality with real‑time constraints; an AUC lift of 1.5 % is reported.

Finally, the article presents Baidu’s GRAB (Generative Ranking for Ads) model, which replaces traditional DLRM feature engineering with an end‑to‑end generative sequence model based on 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 authors also describe the STS two‑stage training algorithm, heterogeneous token representations, dual‑loss stacking, and KV‑Cache optimizations that ensure high‑throughput online inference. Quantitative business benefits after full deployment are highlighted, and readers are directed to the original sources for full architecture diagrams and comparative analyses.