How would you design an infrastructure to serve a highly requested ML model with strict latency requirements at Berkeley Research Group?
