Dealing with a messy prescriptive monolith. How do you survive this? [D]

Months ago, I got my first maintenance project. Before this, I had only built new solutions from scratch and maintained my own code. But maintaining someone else's system feels completely different. ​ ​It’s a prescriptive recommendation system that uses XGBoost models and Differential Evolution for optimization. The problem is that everything is in a single repository: raw data ingestion, transformations, model training, reporting, the optimization engine, post-processing, and MUCH more. The only thing outs