Recent years, we’ve been working myself on a vast variety of infrastructures to build up basis elements on which ML enforced apps are delivered under a secured, cost efficient and well-controlled manner.
The following diagram shows a part of (conceptual) single resource unit of my pure private use (this blog belongs to App Server
, here).

You see that there contains no complex topology here, but it turned out a drastic change in the ML Workload Server, which forced me months of work to have it settled as it’s presented.
Taking advantage of the constructed environment, hopefully admit some MLOps-compliant utility, we have experiments and applications in mind that we’d manage to perform in next few months.