How JD’s Tech Teams Power 618: AI, Logistics, and Voice Innovations

Time flies, and the annual 618 shopping festival is back, showcasing how JD’s technology improves user experience through faster logistics, smarter recommendations, and AI‑powered services.

Retail – Changlin: Making Models Affordable and Effective

JD’s “same‑product identification system” groups similar items to help users compare prices, driven by a model that compares product attributes. Changlin, a PhD from the Chinese Academy of Sciences, leads its optimization using model distillation, extracting knowledge from large models into smaller, cost‑effective ones. He also designed a data‑selection mechanism that prioritises high‑information “fuzzy” samples, saving 40‑60% of training resources while preserving accuracy. The method is now used in system auditing.

Logistics – Xingyan: Mastering a Single Scenario

Xingyan heads a ten‑person logistics tech team. He rebuilt the courier‑assignment model by creating an intelligent partitioning system that combines courier profiles (delivery volume, elevator usage, return frequency) with community terrain information, enabling more balanced manpower distribution. He also leads the deployment of robotic arms in sorting centers, focusing on the “small‑package sorting and stacking” scenario and iterating a feedback loop that quickly relabels mis‑identified samples for model retraining.

AI – Chuxue: Voice Recognition at Scale

Chuxue, a PhD‑trained speech researcher, works on JD’s voice‑recognition stack, covering VAD, speaker verification, and large‑scale ASR. He advanced the system from low‑level modules to end‑to‑end projects such as intelligent outbound calls and meeting transcription, achieving performance that surpasses external competitors. To handle diverse accents, his team collected nationwide speech samples and applied a Mixture‑of‑Experts (MoE) architecture for dynamic dialect switching. They also explore perception‑level innovations, analysing emotion and prosody to infer user intent, aiming to make AI truly understand human speech.

Technical talent must focus on long‑term, business‑driven problems rather than chasing algorithmic tricks.

The three engineers share a common mindset: “practical, daring, and grounded.” JD supports them with mentorship, unlimited salary based on ability, and the TGT (Tech Genius Team) program, which recruits top‑young talent worldwide, offers three‑mentor guidance, and provides abundant data and compute resources for research‑to‑production pipelines.