What is the capstone?

Your capstone is the one project you choose and build across five live sessions - not five unrelated homework repos. You pick it: a tool, assistant, or workflow in a domain you care about. Each session adds production engineering to your project: Session 1 adds reliable reasoning (`/ask`), Session 2 grounds it in your data (RAG), Session 3 gives it agency, Session 4 adds TRACE evals, Session 5 adds memory - then you deploy and demo it.

Why we do it this way

Disconnected weekly exercises teach tools. One capstone teaches engineering: how layers stack, how failures compound, and how to make a non-deterministic system dependable enough to ship. You finish with your project demo-ready - something you can show in three minutes, put on GitHub, and discuss in an interview - not a folder of tutorial clones.

Pick a project you actually want to exist. It should do something useful with information or tasks you care about - company docs, tickets, logs, research in your field, customer FAQs, internal wikis, or another dataset you can access. The course teaches reliable patterns (reasoning, RAG, agents, evals, memory); your capstone is where you apply them to your idea.

Your one-page seed (four questions)

1. What does it do? One sentence. If it takes more than one, narrow it.
2. Who is it for? Name the actual person who would use it.
3. What is the single most important question or task it must handle well first? This becomes your Session 1 `/ask` domain.
4. What would make you proud to demo on Demo Day? That is the north star for scope.

Include your capstone one-liner in your pre-session check-in post. If you write the full page, link to it (Google Doc, Notion, or gist) when you share.