Pretty much every developer, when brought in as a replacement, will bad-mouth their predecessor, say their code is junk and needs to be totally rewritten, etc. Managers love it and encourage it, to them it's a divide and conquer tactic. Data scientists just take it to the next level, that's all.
The first thing my colleagues did after he left was replace all the code he wrote in the C codebase (I work on the front-end, there's a few years worth of work yet to rebuild it).
I'm talking stuff like 15K LOC files with 4-5 times nested switch/case/if/else blocks; string concatenating XML to a global variable, then at the end converting it to JSON; string concatenating HTML with embedded JS and CSS in a JS function inside of a PHP file; using a semicolon-separated value in a database column instead of a simple foreign key (more fun when there's an extra value attached to it, separated with a colon).
It's already not an easy domain or a simple application (thousands of fields across dozens of models), but the lack of coding practices only made it worse.