Dream Project: Solve a Problem with AI

Real AI Builders Start With a Problem

Every useful AI tool in the world started the same way: someone noticed a problem and wondered if AI could help solve it. The farmers who built crop disease detectors noticed that inspecting every plant by hand was too slow. The doctors who built cancer screening tools noticed that human reviewers sometimes missed early signs in scans. The engineers who built voice assistants noticed that typing is slow when your hands are full.

You have now built image classifiers, an audio detector, a pose controller, a chatbot, an art generator, and a quiz bot. You know how to gather training data, train models, test them, and improve them. You have a real AI toolkit. Now it is time to use it for something that matters to you.

The best projects solve small, specific problems — not big vague ones. "Help people recycle better" is too big. "Tell me if this item belongs in the blue bin or the green bin" is a project you can actually build. Start small and specific.

Six Project Ideas to Inspire You

If you are not sure where to start, here are six ideas that can each be built with the tools you have already learned:

• Mood Detector: Train an image classifier that detects your facial expression — happy, neutral, or frustrated — using Teachable Machine. Build a Scratch journal that adds the matching emoji when you look at it each morning.
• Plant Health Checker: Take photos of plants around your home or classroom. Train a model to tell the difference between a healthy plant, a thirsty plant (drooping leaves), and a sick plant (yellow or spotted leaves). Useful for anyone who forgets to water things.
• Pet Identifier: Train a classifier to tell your pet apart from other animals, or to tell different dog breeds apart. How many photos does it need before it gets it right consistently?
• Sound-Triggered Story: Train an audio model with three sounds. Each sound makes something different happen in a Scratch story — a new character appears, the scene changes, or a sound effect plays. Build a story where the audience controls what happens by making noise.
• Instrument Recognizer: Can AI tell the difference between a guitar strum, a piano key, and a drum hit? Train a Teachable Machine audio model with recordings of different instruments and see how accurate it gets.
• Body Fitness Tracker: Train a pose model with three exercise positions — arms up, arms out wide, and squat. Build a Scratch counter that tracks how many reps of each exercise you complete based on your poses.

Plan Before You Build

Before you start collecting training data or opening any tool, spend five minutes planning on paper. Write down:

• The problem: What specific thing are you trying to solve or improve?
• The categories: What will your AI need to recognize? List all the classes.
• The tool: Which fits best — Teachable Machine image, audio, or pose? Or Machine Learning for Kids text?
• The output: What happens when the AI makes a prediction? What does the user see?

Planning like this prevents you from getting halfway through building and realizing your idea does not quite work the way you expected.

Build, Test, and Improve

Now build it. Collect your training data. Train your model. Test it. Find what breaks. Add more training examples. Retrain. Test again. Remember everything from earlier lessons: diverse training data beats a big pile of similar examples. Testing with someone else reveals problems you cannot see yourself.

When you have something that works — even if it is not perfect — you are ready for the final lesson: sharing it with the world.