How would you design model serving infrastructure to minimize inference latency while scaling to high traffic at Spring Health?
