What was the orchestration system used for? Was it in the case where there were many models that needed to be run one after another. I know it's a huge problem in video processing to be able to increase speed a ton. My company Sieve (see profile) is building infrastructure specifically for running ML models on video which is why I'm curious.