R, RStudio, and Shiny are great for prototyping, but you'd regret building a production data science ecosystem around them. Better to start with Python and rely on JavaScript viz tools in the end.
In addition to being a language that does almost nothing to encourage good development practices and discourage bad ones, compared to Python the R ecosystem's tooling is just so inferior in every area: IDE, unit/functional testing, ETL, deployment, dependency management, etc. etc.