Predict Stock Return 30 Minutes Ahead

Design features to predict a given stock's return 30 minutes ahead after removing the systematic (market) component.

Asked in the phone screen stage. This is a business-scenario question, and the rest of the prompt is points-walled.

Address:

1. Define the target after removing market exposure and avoid look-ahead leakage.
2. Design point-in-time features and an offline-to-online data pipeline.
3. Choose a model and explain batch versus online inference.
4. Define offline and online evaluation, monitoring, and failure handling.
5. Explain how you would handle feature drift and training-serving skew.