The Genius Project 2026 Results: 200+ Students, 1M in Prizes

Two hundred and something young people signed up. The youngest were five. The oldest were eighteen. They came from across the Caribbean, some sitting in a room in Kingston, most joining from wherever a laptop and a connection would let them.

Over one month they learned to use AI tools properly, then went past the tools into the mathematics underneath them. By the end, teams were presenting working machine learning models for problems their own communities live with. 1M in cash and prizes was given away over that month, funded entirely by sponsors, with no tuition charged to a single family.

This is the record of what happened, what the numbers say, and who made it possible.

What the Students Actually Learned

The curriculum had a deliberate order to it. Tool use came first, because it is the fastest way to give a young person a result they can see. Then the ground shifted under them on purpose.

Students moved from prompting a model to understanding what a model is: training data, features, labels, and the difference between a system that has learned a pattern and a system that has memorised an answer. That meant statistics, and statistics meant mathematics. Averages, spread, probability, and the reason a single result tells you almost nothing. For the older groups it meant Python, notebooks, and the ordinary frustration of code that runs but returns the wrong number.

Two subjects ran through every session regardless of age. The first was teamwork, because nothing in the programme was built alone. The second was problem definition, which is harder than it sounds and is the skill most adults in technical careers are still short of. A team that cannot say clearly what problem it is solving cannot build anything worth judging.

What They Built

Teams were pointed at real problems rather than tidy exercises. The work clustered into four areas.

Crime and community safety. Teams looked at where incidents cluster, how reporting gaps distort what the data appears to say, and what a model can and cannot responsibly predict about a place or a person. Several groups arrived at the conclusion that a predictive model on incomplete crime data mostly predicts where the reporting is, which is exactly the lesson we hoped they would find themselves.

Poverty and access. Work here covered household budgeting tools, food price tracking, and matching people to services they qualify for but do not know about. The constraint the teams kept running into, that the data for Caribbean households is thin and scattered, is the same constraint professional teams hit.

Sport. Football and track data turned out to be the best on-ramp to machine learning we have. Match outcomes, expected goals, race splits and injury patterns gave students a domain where they already had intuition and could immediately tell when a model was talking nonsense.

Ethics and responsible AI. This was not a lecture module. Teams had to state who their system could fail, what data it should never hold, and what they would tell a person the model got wrong. A fourteen-year-old explaining why her model should not be used for hiring is a better argument for AI ethics education than any policy paper.

Across all four areas, students built actual machine learning models. Not slide decks about models. Models that took input, produced output, and could be shown to be wrong.

The Parents Trained Too

A child who understands AI better than the adults around them is not a safe arrangement. So parents ran their own track.

It covered the practical safety layer first: account settings, privacy controls, what a chatbot retains, what to never paste into one, and how to recognise a website that exists to harvest information. Then it went to the harder part, which is judgement. How to tell a generated image from a photograph. How to check a claim before forwarding it. How to recognise AI slop, the fluent and confident text that is wrong, and why the fluency is the trap.

The target was simple. A parent should be able to sit beside their child, look at what the child is doing with an AI tool, and ask a question that improves it.

The Hackathon

The month closed with the final hackathon. Teams presented to judges, defended their build, and answered for the choices they made. Congratulations to the winners, and to every team that stood up and presented. Presenting a technical build to a panel of adults is difficult at thirty. Several of these presenters were not yet thirteen.

The Number We Are Not Hiding

Completion across all areas currently stands at roughly 15 percent.

We publish that figure because the alternative is to publish only the flattering ones. Large open online programmes commonly report completion in the mid single digits, so 15 percent across a month-long technical programme spanning several countries is a result we are willing to stand behind. It is also nowhere near where we want it.

We know where the drop-off happens. It happens when the material stops being tools and starts being mathematics, and it happens hardest for students without a reliable connection or a quiet room. Both are fixable. The 2027 design work is aimed at exactly that: shorter modules through the mathematics section, offline-capable materials, and a check-in for every participant who goes quiet for more than three days.

Start Early. Start Honest.

The case for starting at five is not that a five-year-old should be writing code. They should not. A five-year-old sorts objects, counts them, notices a pattern, and learns that the computer's guess can be wrong. That last one is the whole thing. A child who learns at five that a machine can be confidently wrong grows into a teenager who checks, and into an adult who does not forward the fake.

The Caribbean cannot wait for AI literacy to arrive through the university system. The students who will be building and governing these tools in 2035 are in primary school right now. Teaching them responsible use early is cheaper, faster and more durable than trying to correct habits later.

Who Paid for It

The Genius Project charges families nothing. That only works because organisations across the region put in cash and support. For 2026 that included:

• StarApple AI, the Caribbean's first AI company, which funded prize money and supplied instructors and curriculum.
• Maestro AI Labs, which contributed technical mentorship and lab time for the machine learning tracks.
• The Caribbean AI Association, which backed the programme regionally and opened doors across CARICOM.
• 14West, which supported the hackathon and the prize pool.
• AI Trinidad and Tobago, which carried the programme to students across the twin islands.
• Orbital Brand Science, which contributed in-kind support and helped the message reach families.
• Adrian Dunkley, founder of The Genius Project, who backed the programme personally and taught in it.

To every sponsor, judge, volunteer instructor and parent who gave up a month of evenings: thank you. To the students: you did the hard part.

Next Cohort

Registration for the next cohort is open at beagenius.org/register.html. It is tuition-free, it takes students from age five, and it does not require any prior coding experience. Programme details are at the Genius Bootcamp page.