Overview
- Discrepancy in how data is handled during training and security predictions
- e.g. Difference in pipelines—maintained by different teams
- Use same code for training and prediction
- Need to handle both batch (training) and streaming (predictions)—Dataflow
- Common reason ML projects fail