Write a text parser and fix the bugs in an ML project

In a one-hour round on an existing ML project: first write a parser that turns raw text data into the format the project needs, then find and fix three bugs elsewhere in a fairly large codebase. Speed at locating where things go wrong matters.