Work · Case study
A catalogue thatdescribes itself.
01 / Challenge
Half the catalogue had no usable description and the other half had three. Search returned the wrong products because the text underneath them did not say what they were.
02 / Approach
We built a pipeline that reads whatever a product already has — supplier feeds, images, specifications — and drafts consistent copy against the client's own tone and taxonomy. Attributes are extracted as structured fields, not just prose.
Nothing publishes unchecked. Quality gates reject anything that contradicts the source data, and a sample of every batch goes to a person before the batch goes live.
03 / Outcome
Products enriched
Better search hit rate
The catalogue describes itself in one voice, and search returns the right products because the words underneath them finally match what is in the box.

