Biophysics grad student dropout/work at startup now; I personally was sort of sick of mathematical modelling, so I decided not to go the machine learning route. But as I was based in Boston, there were machine learning jobs everywhere. I enjoyed working through SICP, and its functional style been pretty useful in quickly understanding new concepts (Javascript callbacks and promises, for example) in programming.

I got my programming start in a website for a tournament that I ran - it taught me my way around a large framework (Django) and as django was what people call a 'highly opinionated and bloated' framework, it taught me one version of how experts think a large-scale project 'should' be organized.