A platform for Zhuanzhuan R&D and industry peers to learn and exchange technology, regularly sharing frontline experience and cutting‑edge topics. We welcome practical discussions and sharing; contact waterystone with any questions.
This article details a step‑by‑step investigation of an OutOfMemoryError in a Spring Boot social app, revealing that frequent initialization of a script engine caused massive StringSequence instances via SPI loading, and shows how consolidating the engine eliminated the OOM issue.
This article reviews the role of pre‑ranking in multi‑stage recommendation pipelines, compares dual‑tower and fully‑connected DNN models, discusses negative and positive sample selection strategies, and presents Zhuanzhuan's practical improvements in model architecture and traffic‑pool allocation to boost precision and diversity.
This article details the engineering practice of optimizing TorchServe‑based AI inference services, covering background challenges, framework selection, GPU‑accelerated Torch‑TRT integration, CPU‑side preprocessing improvements, and deployment on Kubernetes to achieve higher throughput and lower resource consumption.
This article examines the characteristics of ZhaiZhai's B2B auction scenario, defines core pricing metrics, presents a step‑by‑step methodology for determining optimal starting prices, reviews early practices and their shortcomings, and details the current modular machine‑learning model architecture that improves transaction rates and reduces price premiums for second‑hand smartphones.
This article introduces multi‑task learning, reviews industry category‑prediction methods, and details Zhaozhuan's practical application of MTL to improve e‑commerce search intent understanding through hierarchical category, brand, and model prediction using RoBERTa and contrastive learning.
This article introduces causal inference concepts, explains structural causal and potential‑outcome frameworks, presents the MACR model for debiasing popularity in recommender systems, and details two experiments conducted on the ZhaiZhai platform along with future research directions.
The article reflects on the difficulties of turning abstract design concepts such as GOF patterns and DDD into practical, reusable middle‑platform services for an e‑commerce business, discussing when to abstract, the trade‑offs between thin and thick services, and how to balance efficiency with evolving business needs.
This report analyzes performance bottlenecks caused by serialization in a search‑recommendation system, presents detailed measurements of request latency, evaluates multiple optimization strategies—including Redis caching, lazy metric handling, and custom byte‑array serialization—and documents the resulting latency reductions and implementation considerations.
This article explains how ZuanZuan calculates and displays expected delivery times, outlines the system architecture across order, OMS, WMS, and TMS modules, and describes caching and high‑concurrency techniques used to support millions of delivery‑time queries per second.
This article examines whether the REPEATABLE‑READ isolation level in InnoDB fully prevents phantom reads by conducting systematic current‑read and snapshot‑read experiments, analyzing lock mechanisms, MVCC behavior, and presenting detailed SQL scripts, results, and recommendations for avoiding phantom reads.
