What I Read: Improving a Machine Learning System (Part 1 – Broken Abstractions)