Awesome! I’ll reach out next week! 
We’ve been handling FHIR mostly “as needed” as well. I wrote a library that was built using the specs from versions 2 thru 4… and would move data to_json or to_fhir or to_map – but after doing so… discovered some issues. I had used their XML schema stuff – only to find out that it contained perhaps 90% of what’s needed. I don’t know why they bother linking it and other document types when they’re apparently incomplete. Further, Elixir’s Jason library doesn’t support roundtrip decimal precision preservation – which is required for FHIR. I ended up creating a local fork of Jason to do so after failing to get the author to support it (note: it appears that his mindset has changed and there are discussions now to support decoding strategies).
At some point, I’m hoping to go back and build an open source Elixir library for FHIR based on what they call their “structure definitions” – which are supposedly the only documents that are canon and complete. But alas, they’re an awful mess that requires mixing old data designs with new… it has to be one of the most dismal specs I’ve ever seen in terms of data definitions.
