Understanding and Optimizing Multi-Stage AI Inference Pipelines | Read Paper on Bytez