Deploy a real-time SageMaker endpoint, benchmark latency at p50/p95/p99, and measure cold start. This project costs roughly $0.056/hour while the endpoint is up — it ships with a teardown script; run it the moment you are done testing. This is the live endpoint the next two projects (A/B testing, monitoring) build directly on top of.
Benchmarked p50/p95/p99 latency on a real SageMaker endpoint, including cold start, is the exact metric a hiring manager checks for when screening for production ML experience.