Skip to content

SYSTEM Cited by 1 source

Zalando Article Masterdata

What it is

Article Masterdata is Zalando's internal system-of-record for article-level metadata — the catalog of attribute codes, attribute-set definitions per article type, and the human-readable English labels used for LLM prompting. It is the authoritative source for "which attributes this article type has", "what values each attribute can take", and "which attributes are mandatory vs. optional" (Source: sources/2024-09-17-zalando-content-creation-copilot-ai-assisted-product-onboarding).

Role in the copilot workflow

The Prompt Generator pulls three things from Article Masterdata per suggestion call:

  1. The article's attribute set — which attributes apply to this article type, with their codes (e.g. assortment_type, possible values assortment_type_7312, assortment_type_7841).
  2. English translations of the codes — so the LLM can see Petite and Tall instead of opaque identifiers. This is the load-bearing input to the concepts/opaque-attribute-code-translation-layer.
  3. Mandatory vs. optional flags — feed the concepts/category-attribute-relevance-mapping so that irrelevant attributes for this article's category are suppressed from the prompt entirely.

Architectural properties

  • Codes, not names, are the canonical identifier. "We are storing the attributes of the article as attribute codes" — the LLM never sees these codes directly. The translation layer is purely for the human-readable path in and out of the model.
  • Multi-market via codes. Because attributes are stored as codes, the same article has one canonical record across Zalando's 25 markets; per-language rendering is a display-layer concern.
  • Schema source-of-truth for prompt materialisation. The Prompt Generator is schema-driven — it reads which attributes exist and what values they can take from Masterdata, not from a hand-maintained prompt template. New attributes become available to the copilot the moment they're defined in Masterdata + given English labels + not excluded by the category-relevance map.

Relation to Zalando MDM

This is not the same system as the Zalando MDM system described in the 2021-07-28 post — that one is a consolidated-style master-data-management component for internal data assets across multiple source systems, and its design uses a Neo4j knowledge graph to generate the logical data model. Article Masterdata as described in the 2024-09-17 copilot post is a narrower catalog metadata service specifically holding article attribute codes and attribute sets. The posts don't cross-reference — they may or may not share infrastructure.

Seen in

Last updated · 501 distilled / 1,218 read