How do you improve integration when working with Machine Learning?

[elliotthinker64]

I have a question about Machine Learning that may be useful for other people working with it too.

I am particularly interested in integration and accessibility. I would like to compare approaches that people have actually used rather than just theoretical suggestions.

What has worked best for you, and what would you avoid?