An AI cataloguing system that knows when it’s unsure.
the short version
An in-house AI cataloguing pipeline that tagged 8,000+ SKUs across seven attributes, with confidence scoring and human-in-the-loop review.
La-Fetch · Founding Product Manager · Jun 2025 — Present
a sketch of it
fig. 1the flow, as I’d draw it on a whiteboard
1. what i built, what i changed
I led an in-house AI cataloguing system that tagged 8,000+ SKUs across seven attributes, combining product descriptions, existing tags and images.
Tags came from a consensus of two models, each scored high, medium or low in confidence. High-confidence tags were approved automatically; about 1,500 uncertain SKUs were routed to human review.
2. the outcome
The resulting attribute tree became the site’s navigation taxonomy.
what I learned
not written down yet. it’s a better conversation than a paragraph, anyway.