Skip to content

Capstone: build & demo a complete AI app

This is the finale: design, build, and demo an AI app that's yours. It's not a new skill, it's combining what you already did in M4-M10 into one thing that runs and that you can show off.

Pick a track

Track What you build Reuses
Knowledge assistant (RAG) Q&A over a real document set you care about, evaluated and deployed. M7 + M8 (+ M11)
Action agent A tool-using agent that completes a multi-step task, with guardrails. M9 + M10 (+ M11)
Your idea Anything LLM-powered that solves a real problem from your life or work. any of M4-M10

Requirements (the bar)

Your capstone must: 1. Run: a working app (a script, or served via FastAPI / Docker from M11). 2. Handle a basic failure gracefully: e.g. a missing file, a bad input, an empty result, or a blocked/guardrailed request. It shouldn't crash on the obvious thing that goes wrong. 3. Be explainable: you can say, in your own Python, how it works and how you'd secure it (which guardrails, what could be attacked, what you'd add, M10).

How to build it (start small)

  1. Write one paragraph: what it does, for whom, and the smallest version that runs.
  2. Get the smallest version running first (a single call, hard-coded input). Then add the real input, the RAG/agent, the guardrail, the API wrapper. Working-then-better beats perfect-then-never.
  3. Reuse your modules: drop your M7 RAG or M9 agent into starters/app_starter.py's answer_question(), or run it as a script.
  4. Add one guardrail and one graceful failure (M10, M3's try/except).
  5. Secrets hygiene: key in .env, never committed, passed at run time if deployed (M4, M11).

Demo checklist (~3-5 minutes)

  • Show it running and do one real example live.
  • Trigger your graceful failure once (show it doesn't crash).
  • Say how it works in one breath (retrieve→augment→generate, or the agent's tools), and one thing you'd do to secure it.
  • Take a bow.

Go further (optional)

The AI Engineering Resource Map points to deeper resources for every piece, RAG evaluation, agent frameworks, the security track, inference optimization, and hosting options, when you want to take your capstone past the course.