Choose a small project with inspectable inputs, a clear test and a component the learner can change. Our recommended first goal is a defensible prototype, not a public service or a claim of professional-level research.
Explore the comparison in this briefing
Project idea
Hobby-item classifier
- Bounded first version
- Classify a small, cleared dataset
- Learning evidence
- Explain training examples and investigate wrong predictions
- Keep outside the initial scope
- Photos of classmates or family members
Compare each option against the same criteria. The complete table and qualifications remain in the briefing.
These are desk-designed examples to scope with the mentor:
| Project idea | Bounded first version | Learning evidence | Keep outside the initial scope |
|---|---|---|---|
| Hobby-item classifier | Classify a small, cleared dataset | Explain training examples and investigate wrong predictions | Photos of classmates or family members |
| Fictional-club FAQ assistant | Answer from a short invented handbook | Identify unsupported questions and test refusal behavior | Real school records or public users |
| Sports-statistics explorer | Compare a simple baseline with a small model | Explain the split between development and evaluation data | Betting recommendations or performance promises |
| AI-answer checking project | Compare answers with approved source material | Document errors and explain verification steps | Treating chatbot answers as authoritative |
For a trained model, reserve evaluation data appropriately rather than repeatedly adjusting the model to the same test examples; scikit-learn explains why this separation matters. (scikit-learn.org)
For a 12-year-old, the account plan needs particular attention. GitHub’s published terms require users to be at least 13; do not assume every common development platform is available to the learner. Use a permitted educational or local workflow instead. (github.com)
Ask the child which problem interests them before selecting the technology.