MedScan Health needs a SageMaker endpoint that runs an image preprocessing container, then a TensorFlow model container, then a postprocessing container in sequence for each request. They want a single HTTP endpoint URL for clients. Which SageMaker deployment approach best satisfies this requirement with minimal extra networking and correct model lifecycle behavior?

The correct approach is to create an inference pipeline because it directly supports chaining multiple containers (preprocessing, model, postprocessing) into a single SageMaker Model. This model is then deployed to a single endpoint, providing one HTTP URL for clients and ensuring SageMaker manages the sequential execution and lifecycle of all containers. Deploying three separate endpoints with an ALB/Lambda adds unnecessary networking complexity and latency. A multi-model endpoint is designed for hosting multiple independent models on shared infrastructure, not for sequential processing within a single request. SageMaker Processing jobs are for batch processing, not real-time inference, and would not meet the low-latency, single-endpoint requirement.