JD Retail Data Platform: Data Asset Governance, Metric Middle Platform, Visualization Tools, and AI‑Driven Applications

JD Retail’s self‑operated and merchant‑operated models generate massive, complex data dimensions, making data retrieval and storage‑compute cost reduction a core challenge for the data team in 2023.

Through a year of iteration, the team accelerated indicator development and sharing, shrinking dashboard build time from days to hours and growing daily indicator consumption from millions to tens of millions.

The article is organized into four parts: (1) Data Asset – certification and governance, (2) Data Capability – metric middle‑platform practice, (3) Data Presentation – visualization tools, and (4) Data Intelligence – large‑model‑based intelligent applications.

Data Asset Certification standardizes assets, enforces completeness and uniqueness, and retires redundant assets, covering core retail domains such as transaction, user, traffic, marketing, and finance.

Data Asset Perception provides end‑to‑end visibility through automated asset graph construction, enriched model detail pages, and a standardized field library, improving asset discoverability and usability.

Metric Middle‑Platform addresses challenges of metric ambiguity, resource gaps, and sharing difficulty by offering full‑cycle indicator asset control, a native topology market, rule‑engine‑driven formula management, proactive anomaly alerts, and logical‑wide acceleration via materialized views and engine selection.

The platform’s architecture follows MDA and DDD principles, separating physical, semantic, and query layers, with unified DSL for query expression. Example DSL:

{
    "indicators": ["ge_deal_standard_deal_ord_amt"],
    "attributes": ["shop"],
    "criteria": {
        "criterions": [
            {"propertyName": "main_brand", "values": "8557", "type": "string", "op": "="},
            {"propertyName": "dt", "value": "2023-12-21", "type": "string", "op": "="}
        ],
        "orders": [{"ascending": false, "propertyName": "ge_deal_standard_deal_ord_amt"}],
        "maxResults": 5,
        "firstResult": 0,
        "group": ["shop"]
    }
}

The query engine performs semantic and engine splitting, generating DAG‑based execution plans that reduce redundant data scans and improve TP99 latency.

Visualization Tools leverage graphic‑grammar theory to build flexible components such as DuPont analysis, grid metric cards, and cross‑analysis tables, supporting PC and mobile, low‑code composition, and automated report generation.

Low‑Code Orchestration provides MVC‑based state management, visual composition, data‑set orchestration, code generation, and injection, enabling rapid page building, multi‑device deployment, and emergency response capabilities.

Data Push automates email delivery of dashboards by rendering canvas snapshots via Chrome DevTools, handling cross‑platform rendering and image generation.

AI‑Driven Data Intelligence introduces LLM‑based natural‑language query, prompt engineering, and local model fine‑tuning to translate user intents into structured queries, improving accuracy through exact, similarity, and behavior‑driven matching.

The platform delivers significant business value: over 40 million daily data calls, support for 8 000+ indicators, 22 data products, a 70 % increase in delivery efficiency, and accelerated AI‑enabled analytics for decision‑making.