AI Explorers · lesson 22 of 24 · 25 min
You’ve learned how AI senses, reasons, learns, talks and affects people. Now it’s your turn to design an AI helper for your own community. Over the next three lessons, you’ll take it from idea to pitch.
The capstone plan
Your capstone follows the same steps real AI teams use:
- Problem: a real need in your community (this lesson).
- Data: what examples the AI would learn from (this lesson).
- Model: a working prototype in Teachable Machine (lesson 23).
- Test: measure how well it works (lesson 23).
- Fairness check: make sure it works for everyone it should serve (lesson 23).
- Pitch: present your idea clearly (lesson 24).
Keep everything in one notebook or folder. That is your project journal.
Step 1: find a problem worth solving
Good AI projects start with people, not technology. Think about places you know: your block, school, church, a family business, a rec center, a barbershop, a community garden.
Ask:
- What is annoying, slow, confusing or unfair there?
- Who has the problem? Talk to at least one of them.
- Would AI actually help, or would a simpler fix work better? (Sometimes a sign or a checklist beats an AI.)
Example ideas to get you thinking (choose your own):
- A sorting helper that recognizes recyclable vs. trash items at a school cafeteria.
- A sound helper that recognizes a smoke alarm or doorbell and flashes a light for someone who is hard of hearing.
- A garden helper that tells healthy leaves from leaves that need attention.
- A practice helper that recognizes correct and incorrect hand positions for a sport or instrument.
Write a problem statement in this form: “[Who] has trouble with [what] because [why]. An AI helper could [how].”
Step 2: plan your data
Your helper must be something you can prototype with Teachable Machine: an image, sound or pose model. Plan:
| Question | Your answer |
|---|---|
| What are the classes? | Example: “recyclable,” “trash,” “nothing” |
| How many samples per class? | Aim for 40 or more, balanced |
| What variety do you need? | Lighting, angles, backgrounds, different items |
| Who or what is in the data? | Objects only? Voices? Get permission for any person |
| What could be a hidden shortcut? | Example: all recyclables on a blue tray |
| What will you keep aside for testing? | Items or recordings not used in training |
Do not worry if your first idea feels too small. Small, real problems make the best projects, because you can actually test them. A helper that sorts recycling in one cafeteria is more useful than a vague idea to fix the whole world. You can always grow the idea later, once you know it works.
Try it
- Brainstorm at least 5 problems in your community. Circle the best one for an AI helper.
- Talk to one person who has that problem (a family member, teacher, coach or neighbor). Write down two things they told you.
- Write your problem statement using the template above.
- Fill in the data plan table in your project journal.
- List the materials you need to collect samples (objects, a quiet room, permission from people).
Check yourself
Why should an AI project start with people and not with the technology?
The goal is to solve a real need. Talking to people shows whether the problem is real and whether AI even helps.
What four parts go into the problem statement template?
Who has the problem, what the trouble is, why it happens, and how an AI helper could help.
Why plan test data before you train?
So you have new examples to fairly check the model, instead of testing on what it already saw.
Key words
- Capstone: a final project that brings together everything you learned.
- Problem statement: a short description of who has a problem and why.
- Prototype: an early, simple version of an idea you can test.
- Project journal: your notes, plans and results in one place.
Previous lesson: AI, jobs and who gets to build it
Back to AI Explorers
Next lesson: Capstone part 2: build, test and fairness-check