LLVM's relationship to machine code is a bit like Deep Learning's relationship to Bayesian models. Deep learning abstracts out details into a reliable black box, unlike Bayesian models in which all details need to be known. Likewise, LLVM abstracts out machine code in a reliable fashion, so that higher level programming languages don't have to think about that stuff.
I send out a monthly newsletter with tips and tools for data scientists. Come check it out at TinyLetter.
If you would like to receive deeper, in-depth content as an early subscriber, come support me on Patreon!