What were the computational bottlenecks in your previous machine learning pipeline, and how did you optimize them?
