Part 4 · Lesson 21 · Expert

Capstone: The AI Product Stack

Watch one prompt cross every layer you have learned.

In this interactive Infrastructure lesson from Stencel Academy you don't just read — you type real commands like demo request "explain DNS", loadtest 1000rps and watch the machine respond on a live, animated map. It is part of the free How Infrastructure Works course.

Key takeaway: Every lesson was one hop of this map — a prompt architect designs the whole board: latency, cost, scale.

Before this lesson

Builds on: Logs, Metrics & Traces.

What you'll practise

demo request "what is a load balancer?" — A prompt someone else already asked — the cache is shared.
kubectl get hpa promptd — The autoscaler is just a rule you wrote down.
redis-cli -h cache-01 INFO stats | grep keyspace — The cache keeps score of every hit and miss.
demo cost --last-hour — Where the money actually goes in an AI product.

Terms you'll learn

autoscalingcache hit ratioload testrequests per secondcost per request

Check your understanding

At 1000 rps, what cut the LLM bill by ~10×?

The cache answering repeated prompts — 91% hit ratio
The autoscaler adding two more pods
The CDN compressing the responses

Open the interactive lesson →