Agentic system that monitors marketplace listings, detects content drift, and pushes corrected copy and imagery back through Amazon SP-API. Compresses catalog QA from days to minutes.
Marketplace catalogs drift constantly — copy gets overwritten by other sellers, images go missing, variants detach, and listings get suppressed without warning. Teams usually find out after sales dip, and auditing hundreds of SKUs by hand takes days of work that nobody wants to own. The cost isn't just the labor; it's the revenue lost in the window between something breaking and someone noticing.
Instead of a faster audit checklist, I built an agent that never stops auditing. It pulls live listing state through SP-API on a schedule, diffs it against a golden record of what each listing should say and show, and classifies every difference: harmless, needs review, or needs an immediate fix. For fixable drift it drafts the correction itself — copy rewrites stay inside guardrails derived from the golden record — and queues it for one-click human approval before anything is pushed back.
SP-API integration with rate-limit-aware request queueing, so the agent stays inside Amazon's throttling budgets
A diff engine that compares live listing content against a golden record per SKU — title, bullets, description, imagery, and variant relationships
LLM-based drift classification and copy repair via the AI SDK, constrained to the approved content in the golden record
A human-in-the-loop approval queue with full audit history of every detected change and every pushed fix
Scheduled runs on Vercel Cron with state in Supabase — no servers to babysit
Catalog QA that took days of manual spot-checking now runs continuously, and drift gets caught the same day it happens instead of weeks later. Humans only look at the queue of proposed fixes — the searching, diffing, and drafting is gone from their week entirely.
I take on a small number of builds like this each year — agentic systems and tooling for marketplace teams.
GET IN TOUCH