On: How GLM built its own inference infrastructure

[a34729t]

I mean all the database stuff is obvious low hanging fruit for inference engines.