Just curious if you’ve benchmarked your home-grown model against Laya, Kev, etc.
Open-source is tricky these days but I wouldn't be surprised if you got at least one or two useful contributions from others knowledgeable in this space with similar goals.