Start course

Week 1 · Day 1 of 10

Your first AI app, live

Every student has their own AI app answering on a public URL, built with a coding agent, and can explain every line of it.

Day 1 welcomeHi, I'll walk you through today.
Your guide walks you through today: what you'll ship, how a day works, and the seven build steps. About 45 seconds.
In pink: what day 1 adds to the app.
Day 1 adds Your user, FastAPI app, Model to the student's app.
1LearnShip a thin slice with a coding agent

Working like a forward deployed engineer: start from the user's problem, then get the thinnest possible slice live the same day, so access, keys, deploy config and cost problems show up while they're cheap to fix. Using a coding agent well, and the rule that you must be able to explain every line it writes. Calling a model through the AI Gateway, keeping the key out of the code, and deploying. Only the theory the build needs, when it comes up, like what each call costs.

How it worksStep 1 of 5

How your app answers a question

  1. You ask
Someone types a question and sends it to your app's public link.
Why Vercel3 reasons
  • One command to go live

    vercel deploy puts a Python app on a public link with no servers to set up, so day 1 is spent building, not configuring.

  • The model key lives there too

    The AI Gateway gives one key for Claude and other models, with a spending limit, next to where the app runs.

  • Every push deploys

    Once the code is on GitHub, each change goes live by itself, with a preview link to show people.

At a partner company you'll often deploy to their cloud instead, such as AWS, Azure or Google Cloud. The habits carry over: keep the key out of the code, ship a thin slice early, and deploy on every push.

Check your understanding2 questions
  1. Your app answers on your laptop but errors once it's deployed. What's the most likely cause?

  2. A model is 3× cheaper and 2× faster, but slightly worse on your five test prompts. How do you choose?

2BuildPut your first AI app online

Watch Robert put the app online first, one short clip per step, then do each step yourself: someone sends a question, Claude answers. Read every line your agent writes until you can explain it, then show your link to a classmate.

Done when: a link anyone can open gets an answer from Claude.

  1. 1 Check your tools

    Done when uv, Vercel, Git and GitHub all print without an error.

  2. 2 Create the project

    ⌘KFirst, wipe the terminal window clean

    1. uvThe tool that runs your Python project
    2. initStart a new project in this folder
    3. --appIt's an app you run, not a library
    4. --no-packageKeep it simple: plain files, nothing to install
    5. --python 3.13Use Python 3.13
    6. &&If that worked, then…
    7. lsList the files it made

    Done when the placeholder prints "Hello from my-capstone!"

  3. 3 Give your agent the spec

    Done when the agent has written the app, and your key file is ignored by Git.

  4. 4 Make your AI key

    Done when your Vercel project lists AI_GATEWAY_API_KEY for all three environments.

  5. 5 Run it on your laptop

    Done when your app answers a question on your laptop.

  6. 6 Put it online

    Done when your app answers on its public link.

  7. 7 Save it to GitHub

    Done when your code is on GitHub, without your key.

3CapstonePick a user and a problem

Pick a real user and a real problem. Write a one-page brief.

4HomeworkCompare three models on five prompts

Run the same five prompts on three models and log quality, cost and latency.

For teachers: the day 1 teaching kit.