In my iterations of learning to do this (in the open source machine learning space), there have been always five phases:
i) Ingest - use existing tools for years, try to build things and understand the subtleties of their limitations.
ii) Build something - if you were right about the limitations and identify ways to overcome them, people will recognize that and want to use what you build.
iii) Maintain and learn how your assumptions live up to user realities.
iv) Realize very likely that your first attempt at greatness wasn't all that yet, even if it becomes popular. Popularity is not a sign of what you building being well designed, just that the need for it may be great.
v) Build something based on the combination of your initial hunches and taste which got you into it to begin with, and your experience on building something large.
Iterate, and given enough talent and sweat, you might create something that is just right.
Being a computer scientist but not a programming language geek and looking at this, the project looks like a first attempt at ii), but is selling itself as a late iteration of v). Tech communities have very high standards and taste on design, so overselling is deadly.