Grokking the AI System Design Interview
Vote

0% completed

​

Monitoring, Drift, and Retraining

1. The System Is Getting Worse Right Now

An ML system that gets deployed and left untouched gets worse over time. Not because anything actually breaks. Because the world it was trained on keeps changing. Fraudsters test the fraud model and adapt around it. A pandemic changes delivery times overnight. A new slang word makes a toxicity classifier wrong. This decay has a name: drift. Designing for it is what separates "I trained a model" answers from "I ran a model in production" answers.

There are two kinds. Learn to distinguish them by name, because interviewers do:

  1. **Data drift

.....

.....

.....

Like the course? Get enrolled and start learning!