chapter 22 · part 4

Capstone — ship your own AI

the hook

Everything you've built, combined into one project that's actually yours.

the idea

This is it — no new concept, no new playground pattern. Just you, picking a real project, and using everything you've built across the last 22 chapters to actually ship something.

Pick ONE of the 3 briefs below. Each step tells you exactly which earlier chapter taught you how to do it — this isn't new material, it's assembly.

You don't have to build the FULL thing outside this app to get the value — even sketching out each step, referencing back to the chapter that taught it, cements the whole course into one coherent thing you actually understand end to end.

playground — your hands now

what: pick one of 3 real project briefs and check off steps as you go — each step links back to the chapter that taught it.

why: this is the moment 22 separate lessons become one thing you actually built, not 22 things you separately remember.

how: pick a project card, work through the checklist, check off steps as you complete (or plan) them.

start here → pick a project card, then check off steps as you complete or plan them.

RAG chatbot over your own docs — 0 / 7 steps checked

pick a real document you care aboutch15 →chunk it with overlap — write your chunking functionch15 →embed the chunksch15 →build the similarity searchch16 →assemble the retrieval-augmented promptch15 →wire up a real Gemini call for the final answerch13 →test with 5 real questions, note when retrieval failsch19 →

try this

Actually attempt at least ONE step for real, outside this app — even just chunking a real document (ch15) or writing one tool's JSON contract (ch17). the checklist is a guide, not the finish line.

checkpoint

1. Why does the capstone deliberately avoid teaching any new material?

2. Why does each capstone step reference back to a specific earlier chapter?

recap card — keep this

you didn't finish a course. you shipped an AI product.

next chapter → The PyTorch bridge