About this project
A Belgian furniture business sells a supplier's catalog it does not stock itself. That catalog runs to thousands of products and many times that in images, and it changes every day: prices move, stock runs out, new items appear. Doing that by hand is not a job anyone can hold.
So we built the pipeline that does it. Every night the full catalog is pulled in, reconciled, and published to the storefront. Three more passes during the day repair drift: products that lost their stock, their price, or their imagery are quietly taken offline again instead of sitting there as a promise the shop cannot keep. An operator workspace sits on top, where the team can follow every run, see what changed, and intervene where it wants to.
Supplier data is never clean, and a wrong price or a hidden bundle reaches customers as a real problem. Our own rules catch most of it: price floors per furniture category, patterns that betray a set sold as a single item. Where those rules are unsure, an AI model gives a second opinion on the title, the price, and whether the listing is secretly a bundle, and anything it flags is held back from the catalog. It is a gate on top of deterministic checks, not a replacement for them, and it can be switched off without stopping the pipeline.
Highlights
- Thousands of supplier products reconciled into the storefront every night
- Three self-heal passes a day, so broken listings do not linger
- AI-assisted validation flags mispriced and secretly bundled items
- An operator workspace to follow every run and step in when needed

