Our latest work on serving machine learning inference workflows on heterogeneous infrastructures is available here.