I think that set theory and other analytic can be instructive, but can also obscure what's happening. The exact encoding of numbers using sets is just an "implementation detail", in the sense that there's many ways you can build natural numbers (for example) using set theory and they are all equivalent.

So it's like learning data structures by coding in assembly, which is what Donald Knuth thinks is the right thing to do anyway, but some other teachers would disagree. But if you want to see some high level construction, you could look to eg. Tarski's synthetic construction of reals

https://en.wikipedia.org/wiki/Tarski%27s_axiomatization_of_t...

Which doesn't build reals using other theories as building blocks; real numbers are real numbers.

Anyway, from the analytic constructions of reals, I'm most partial to

https://en.wikipedia.org/wiki/Construction_of_the_real_numbe...

Which uses integers rather than sets as the building block, and is simpler than many constructions. And, of course integers themselves can be constructed out of sets, but they can be constructed out of lambda calculus terms as well https://en.wikipedia.org/wiki/Lambda_calculus#Encoding_datat... among many other constructions - but when we finally define integers, we can abstract away the implementation details (and that's really the crux of the question!)

Anyway there's a discussion of analytic vs synthetic mathematics in this post, https://golem.ph.utexas.edu/category/2015/02/introduction_to... (it seems that part 2 wasn't written unfortunately)

(The first half anyway, after that it just becomes a sketch proof of the Banach Tarski theorem)