[sheenobu]

You are right about the input requests being too complex. From this microservices presentation[0], they do use machine learning to generate recommendations which are then curated further by the stylists, specifically for this reason.

But I've always wondered about the grouping and parity between stylist and customer. IIRC, they say it in the presentation but they do not say how much bucketing happens. Is Person of archetype-A grouped with 100 other people and given to Stylist A? 1000 other people? Or do they only do additional curation if you reject so many items and write so many input request words?

Also, is there a good track record for such a request like "I need a pair gold earrings" working? and working specifically because of human curation and not pattern matching on color and item description?

0. https://youtu.be/E8-e-3fRHBw?t=332