Give it a Musinsa product link and one line like “usually Uniqlo M,” and it picks a size to buy. It first looks at what sizes people in the same height and weight range actually bought for that product. For new items with no reviews, it compares measured garment dimensions. Each answer comes with a confidence level: “reliable,” “take another look,” or “for reference only.”
My first version matched measured dimensions directly. Testing showed that actual purchase reviews worked better, so I rebuilt it. People with the same build who bought the same product chose the same size only 66.4% of the time. I treated that as the ceiling for this problem, so it does not claim a single right answer. Code calculates the numbers, and the AI only explains them in plain language.
It is a Codex plugin I built for a company task in the 2026 AX Talent War. The numbers are measured across all collected reviews. They do not measure fit accuracy for any one person.