Official platform of JD Retail Technology, delivering insightful R&D news and a deep look into the lives and work of technologists.
This article compares the storage models, data organization, indexing, read/write processes, and performance trade‑offs of three major message queues—Kafka, RocketMQ, and JMQ—providing detailed technical insights for architects and engineers making storage‑related design decisions.
Uni-Layout introduces a unified layout generation framework that consolidates diverse design tasks, leverages multimodal large language models for flexible generation, and aligns outputs with human perception through a novel human‑feedback dataset (Layout‑HF100k) and a dynamic margin preference optimization (DMPO) evaluator.
The article details JD’s data lake architecture, its 500 PB scale, self‑developed Hudi extensions—including LSM‑Tree‑based MoR tables, custom indexing, IO optimizations, Flink stream scheduling, and NativeIO SDK—along with benchmarks, community contributions, and future roadmap for real‑time big‑data processing.
This article details the complete redesign of a fifteen‑year‑old e‑commerce review system, covering its legacy pain points, the strategic choice of a full‑stack reconstruction using Domain‑Driven Design, the new layered micro‑service architecture, data migration tactics, operational challenges, organizational safeguards, and the measurable performance gains achieved after launch.
This article explains a SIGIR 2025 paper that tackles cold‑start click‑through‑rate prediction in JD's ad system by using a Graph Isomorphism Network‑based cohort modeling framework, detailing its three‑module architecture, extensive experiments on public and industrial datasets, and a live deployment that achieved a 2.13% CTR lift.
This article presents TANDEM, a bi‑level data‑mixture optimization framework that uses twin networks to automatically adjust domain‑specific training data ratios, offering theoretical guarantees, broader applicability, and significant performance gains across pre‑training, fine‑tuning, and e‑commerce product‑understanding tasks.
The JD‑hosted Apache Hudi Meetup showcased the 1.1 release’s pluggable table format, Flink performance gains, LSM‑Tree MoR redesign, and AI‑native features such as vector indexing, illustrating how the open‑source lakehouse is evolving to meet BI and multimodal AI workloads.
The article provides an in‑depth technical analysis of DongSQL V1.1.0, detailing its RETURNING clause, Hint extensions, CCL concurrency control, Statement Outline, single‑point query bypass, thread‑pool redesign, and benchmark results that show performance gains up to 215% in e‑commerce workloads.
An in‑depth post‑mortem of a JED database incident reveals how its lock matrix and MVCC isolation caused metric data loss, explains the underlying lock granularity, transaction isolation levels, and MVCC visibility rules, and proposes short‑term distributed‑lock and long‑term read‑calc‑write solutions.
This article explains the concept of heterogeneous treatment effects, outlines how to select dimensions for HTE analysis, describes a Python‑based MVP tool for automated CATE exploration, and showcases a real‑world experiment case where sub‑group insights turned a non‑significant overall result into actionable findings.
