How would you optimize data pipelines for high-throughput training for Berkshire Grey's machine learning teams?
