The business problem
Every daily product feed arrived without approved taxonomy data. That created a manual categorization backlog that returned with each new batch, but modifying the established importer would have added risk to a time-sensitive production process.
What DSWD delivered
DSWD added a separate enrichment step between the raw feed and the existing importer. It classifies products against a reviewed rule set, stamps the fields the importer expects, and delivers the enriched batch to the location the importer already watches.
The operator experience
Normal batches continue through the existing workflow. Items that match approved rules arrive categorized. Uncertain items appear in a short daily review list, giving an operator a bounded exception queue instead of a full catalog to inspect.
Safeguards and operating qualities
- The import takes priority: if classification cannot complete, the original batch passes through unchanged.
- Partial enrichment is never presented as a complete result.
- Monitoring distinguishes errors, delayed work, and the safety pass-through path.
- New store behavior is configuration-driven and verified by tests.
- Deployment changes are backed up, checked, and reversible.
Project-specific technical choices
For this project, we used event-driven cloud processing and a deliberately narrow file contract because the new capability had to fit inside an established daily import without coupling itself to that importer. The separation protected the critical path and made failure behavior predictable.
Evidence and claim boundary
The service has been live in a daily production pipeline since July 2026. The page describes demonstrated behavior and safeguards, not a promise that every catalog can be classified without business-specific rules and review.