Whenever you're ready, walk me through how you'd approach this.
How do you handle feature engineering for models that require real-time inference?