Guide · Aug 2026 · 1 min read
Choosing Your First AI Research Topic: A High-Schooler's Guide
The best first AI project is small, personal and finishable in twelve weeks. Test every topic against three questions: can you get the data, can you finish, do you care?
Three tests for a good first topic
Can you get the data?
The most common project-killer is data you can't access. Public datasets, school-collectable data, or images you can photograph yourself all pass this test. 'I'll find it later' does not.
Can you finish in twelve weeks?
A finished small project beats an unfinished ambitious one — every time, in every application. Scope for a working demo by week eight, so there is time to break it, fix it and explain it.
Do you actually care?
You will spend three months with this question. Projects rooted in a real interest — your sport, your city, your hobby — survive the boring weeks. Borrowed topics rarely do.
Directions that consistently work
- Image classification with a purpose: recyclable vs contaminated waste, plant disease from leaf photos
- Text analysis close to home: sentiment in local reviews, readability of school materials
- Sound: classifying urban noise, detecting practice mistakes in music recordings
- Prediction with public data: bus delays, weather effects on canteen queues
- Accessibility: sign-language letters, image descriptions for low-vision users
What the finished project should include
- A working demo someone else can run
- A short research report: question, method, results, limitations
- A GitHub repository with a readable README
- A presentation you can deliver in eight minutes
The limitations section matters more than students expect. Knowing exactly where your model fails — and saying so — reads as maturity, not weakness.
