>understanding modern OS and SDE tools for proper implementation of pipeline etc.

Can you provide an example?

How would you filter out garbage from your training data, for example? If you are trying to use someone elses corpus, would it be "from the scratch" then?