Autonomous Catalogue Agent
An agent that checks marketplace listings and prepares corrections to copy and imagery. A person approves changes before they go back through Amazon SP-API. Catalogue QA takes minutes instead of days.
01. The challenge
Other sellers overwrite copy, images go missing, variants detach and listings get suppressed. Teams often find out after sales fall. Checking hundreds of SKUs by hand takes days, while the broken listings continue to lose sales.
02. The approach
The agent reads live listings through SP-API on a schedule and compares them with an approved record for each SKU. It classifies differences as harmless, needing review or needing an immediate fix. It drafts corrections using the approved content, then queues them for a person to approve before publishing.
03. What I built
A request queue that respects Amazon SP-API rate limits
Comparison of each SKU's title, bullets, description, imagery and variants against its approved record
AI SDK calls that classify differences and draft corrections using approved content
An approval queue with a history of detected changes and published fixes
Scheduled checks on Vercel Cron with results stored in Supabase
04. The outcome
Scheduled checks catch changes the day they happen. The team reviews proposed fixes instead of spending days finding differences and drafting corrections by hand.
Want something like this built for you?
I take on a small number of builds each year. Tell me the job and the tools it needs to work with.
GET IN TOUCH