For a denim trucker jacket (a fairly common outerwear in many menswear lines/people's closets) it recommended: One outfit with Sand colored shoes + Khakis (no contrast between shoes/pants...) Two outfits that had no shirts but another jacket (eg a red track jacket...)

Its not all bad, one example outfit generated from a plain white crwwneck sweater was clever - olive flight jacket + orange trousers + white mocassins.

I wonder how you've built the recommendation engine. Is it using a dataset to generate these (maybe include a real photo of the look?) Or is it generating them based on color/style rules? Or (what Id hope) a combination (learn rules from a lookbook dataset that can give general tips then also fall vs spring tips).

Either way, thanks for sharing - this is a fun tool to play with but it feels like I'm searching for needles in a haystack right now.