Product SEO for e-commerce at catalogue scale, with a review queue so nothing publishes unseen.
Switch between how the work ran before, and how it runs now.
In the order it was built, and why each part exists.
A webhook receives product data, an init step normalises it, and routing logic decides whether the item is new or already known.
Product data is fetched from the live store feeds, then a research and prompt-preparation step assembles the context the model will work from.
Google Gemini produces the SEO content, which is parsed and built into final HTML rather than pasted as raw model output.
Results are written to Supabase and surfaced in the dashboard, where the team reviews, approves and publishes. Nothing reaches the storefront unreviewed.
Product SEO at catalogue scale is a volume problem disguised as a writing problem. Two hundred products need two hundred descriptions that are consistent, accurate and worth indexing.
Done by hand it is never finished. Done fully automatically it publishes things you would not stand behind.
A pipeline and a dashboard, deliberately in that order. The pipeline scores products, generates optimised content and stores the result. The dashboard is where a person decides what ships.
The overview shows total reports, completed, needs review, published and weekly activity, with a review queue, per-product scores, analytics and bulk import.
This platform runs our own storefront catalogues. That is the reason the review queue exists: when it is your own shop, publishing something wrong is not a support ticket, it is your listing.
Every client build inherits that constraint.
2
Live stores processed
100%
Reviewed before publish
1
Queue instead of a spreadsheet