JD Retail’s 2024 Tech Innovations: AI, Supply‑Chain Algorithms, and Development

2024 JD Retail Technology Highlights

JD Retail’s technology team released a portfolio of innovations across user experience, supply‑chain optimization, AI‑generated content, low‑code platforms, data‑lake engineering, cross‑platform development, immersive XR, and AI engine compatibility, illustrating a comprehensive push toward smarter, faster, and more cost‑effective e‑commerce.

App Revamp for Better User Experience

The JD APP entered a phased upgrade in the second half of 2024, introducing a clearer "My JD" home section, enhanced "JD Life" service area, a more intuitive message center, higher‑efficiency product detail pages, and smarter search‑recommendation powered by large‑model AI. These changes aim to help users quickly locate desired items and enjoy a "better and cheaper" shopping experience.

Data‑Driven Inventory Selection and Allocation

JD introduced a data‑driven inventory selection and allocation algorithm that balances stock across eight regional distribution centers (RDCs) and their front‑line distribution centers (FDCs). By analyzing historical order correlations and demand forecasts, the algorithm maximizes order fulfillment rates, reduces stock‑out rates, and lowers fulfillment costs. The solution earned the Daniel H. Wagner Prize at the INFORMS 2024 conference and now supports over ten‑million daily AI calls for more than 350 k merchants.

JD AIGC Content Generation Platform

The self‑developed "JingDianDian" AIGC platform provides merchants with high‑quality product images, copy, and short videos at the click of a button. Core innovations include a proprietary text‑to‑image base, ReferenceNet and ControlNet image models, and a multi‑modal product‑understanding model that builds a knowledge base for marketing copy. The platform processes over 10 million AI calls per day, serving more than 350 k merchants and winning the InfoQ 2024 "Best AI Practice" award.

National Subsidy Platform

A low‑code, reusable subsidy‑qualification platform was built to rapidly launch city‑specific subsidy services, handling "old‑for‑new" exchanges, direct discounts, and coupons. Integrated AI‑driven risk modeling intercepts over 99 % of fraudulent users, ensuring subsidies reach genuine consumers.

Data Lake Architecture with Apache Hudi

JD adopted Apache Hudi for its data lake, focusing on I/O performance, feature richness, and ecosystem compatibility. The system supports high‑concurrency writes, linear scalability, and historical data replay across multiple compute engines. The lake now stores 160 PB of data, contributes over 65 PRs to the Hudi community, and reduces storage‑compute costs by more than 12 million RMB annually while achieving minute‑level data freshness.

Taro on Harmony for Cross‑Platform Development

The open‑source Taro framework enables a single codebase to target HarmonyOS, mini‑programs, H5, and React Native. By mapping React DSL to Harmony CAPI, Taro achieves native‑level rendering performance and reduces development barriers for Harmony developers, accelerating JD’s Harmony app launch and cutting R&D costs.

Vision & “立影计划” AR/VR and Naked‑Eye 3D

JD Vision, the first transaction‑enabled Apple Vision Pro app in China, combines spatial computing, eye‑tracking, and hand‑gesture interaction to let users drag 1:1 product models into their real environment. The subsequent "立影计划" introduces tilt‑shift rendering and depth‑culling to create a naked‑eye 3D shopping window, delivering an immersive, high‑fidelity product showcase on mobile devices.

Domestic‑Chip AI Engine

The Nine‑Number Algorithm Platform built a unified AI engine compatible with both GPUs and domestic NPUs. A thousand‑card cluster provides seamless scheduling, zero‑cost model training/deployment, and performance boosts via MFU optimization, quantization, and compilation techniques. The engine now powers multiple JD business scenarios, meeting the growing demand for intelligent services.