Course 9, lesson 90 of 100, Ages 14+

Build your own AI

From idea to working app

Like I’m 5

Building your own AI is like baking a new cake. Pick a recipe, bake a small one, let friends taste it, then make it better.

The big idea

Building an AI product starts with a real problem and a real user, not with the technology. Write down who it helps, what success looks like and what could go wrong.

Then prototype fast: a simple prompt and a handful of test cases. Measure with evals, improve, and only then add complexity like retrieval or agents. Ship to a few users, learn and iterate.

Examples

  • Problem first: 'Teachers spend hours making quiz questions' beats 'let's use AI'.
  • Prototype: A one-page app with a good prompt and ten test cases.
  • Iterate: Each week, review failures and improve the prompt or data.

How it works

  1. Pick a real problem, like sorting recycling or answering questions about your school.
  2. Choose a tool or an AI API, build a small version, and test it with real people.
  3. Measure how well it works, fix what fails, and explain clearly what your AI can and can’t do.

Check your understanding

What should you do after building a first version?
Options: Test it with real people; Never change it; Keep it a secret.
Answer: Test it with real people. Testing with real people shows what works and what needs fixing.
What should you start with when building an AI product?
Options: A real problem and real users; The biggest model available; A logo.
Answer: A real problem and real users. Clear problems lead to useful products.

Remember

Start small, test with real people, improve, and be honest about limits.

Talk about it

What problem around you could a small AI project help with?

Go deeper

A typical project: define the user and how you’ll measure success, prototype with an API or a no-code tool, build a small test set, improve prompts or data, then review privacy, safety and cost before sharing. Write down the limitations honestly.