Kids as Young as 8 Are Building Hurricane Melissa Relief AI Across 5 Countries

The most striking thing in the room is not the technology. It is who is building it. On one of the teams, the youngest member is 8 years old. She is not fetching snacks or holding a clipboard for the older kids. She is arguing, correctly, that a person trapped after a storm should not have to type their full address into a form before the app will help them, because in that moment they may not know exactly where they are and their battery is dying. That is product thinking. It happens to be coming from a third-grader.

This is The Genius Project's newest challenge, and it is built around a name the region already knows: Hurricane Melissa. Instead of teaching AI in the abstract, we handed teams of young people a problem their own families understand in their bones, then asked them to build something real. Across five Caribbean countries, teams are using machine learning and generative AI to solve genuine relief problems and, in plain terms, to get us ready for hurricane season before the next major storm arrives rather than during it.

A New Feature, Built Around a Real Storm

The challenge is a new addition to The Genius Project, and it works differently from a normal lesson. Every team builds its own version of the same core idea: a relief tool that helps people affected by Hurricane Melissa figure out what support they need and connects them with the right help. Teams run a build sprint, produce a working prototype, and present a short demo that is judged on usefulness, reasoning and safety, not on how pretty the interface looks.

That last point matters more than it sounds. A polished screen is not proof that an application works. So before any team opens a builder, they answer a short set of hard questions: who is the user, what is the problem, what information does the app actually need, what decision will the system make, what does the user receive, how could it fail, and where does a human stay in the loop. A team that cannot answer those cannot build. It is the same discipline a professional AI team uses, scaled down to fit a room that includes an 8-year-old and a sixteen-year-old on the same project.

The Five Techniques the Teams Are Working With

This is not one app repeated five ways. Across the cohort, teams are reaching for different pieces of real AI and systems engineering, each aimed at a specific failure that happens during and after a major hurricane. Here is what they are actually building with.

1. Computer vision for emergency rescue

One set of teams is using computer vision, the branch of AI that interprets images and video, to help locate people who need rescue. The idea is to take footage from a phone, a drone or a fixed camera and have the system flag likely people, damaged structures or hazards, so a limited number of responders can look at the highest-priority frames first instead of scrolling through hours of raw video. The AI does not decide who gets saved. It narrows a mountain of footage down to the parts a human should look at now.

2. Machine learning to route relief

Getting supplies to people is not just a driving problem, it is a matching problem. Another group is using machine learning to route needed relief: taking in where requests are coming from, what is being asked for, how urgent each case is and what supplies exist where, then suggesting how to move food, water, medicine and materials so the most urgent, reachable needs are met first. In a real storm, guessing wrong means a truck of water sits in the wrong town while another runs dry. The model's job is to make that guess far less often.

3. Satellite imagery analysed with AI

After a storm, nobody has a clear map of what actually happened. Teams here are using AI to analyse satellite images to assess damage and identify road restrictions: which areas took the worst hit, which routes are flooded or blocked, and where responders simply cannot get through yet. Turning a fresh satellite pass into a usable "these roads are open, these are not" picture in minutes rather than days is the difference between relief that reaches people and relief that gets stuck at a washed-out bridge.

4. Mesh networks for when the towers go down

The cruel part of a hurricane is that it knocks out the very communication people need most. One track is working with mesh networks, where devices connect directly to one another and pass messages along a chain, so a phone can still reach help even when cell towers and home internet are down. A request can hop device to device until it reaches someone who is back online. It is a reminder that disaster tech goes beyond AI: the smartest model is useless if the grid fails and nobody can reach it.

5. Local LLMs with video AI and a knowledge base

The final piece is guidance. Some teams are building local large language models, AI assistants that run on a device without needing a constant internet connection, paired with video AI and a curated knowledge base, to give people in need clear, calm instructions. Think of a resident who needs to know how to purify water, treat a wound, or shut off a gas line, getting a plain-language answer, and where useful a short video, from an assistant that keeps working when the connection does not. The knowledge base keeps the answers grounded in vetted, local information rather than whatever the model happened to memorise.

Safety Is Not the Boring Part, It Is the Point

It would be easy for a young team to build something that looks heroic and quietly does harm: an app that tells a family help is on the way when nobody has actually confirmed it, or that publishes a stranger's home address to the whole internet. So the challenge bakes safety in as a graded requirement, not an afterthought. Every prototype has to carry a clear warning that life-threatening emergencies go to official emergency services. Every match or recommendation the AI makes has to be treated as a suggestion until a human verifies it. And teams have to name at least three ways their tool could fail, each paired with a safeguard.

This is where the challenge doubles as governance training. Data retention, who can see a request, bias in an automated priority list, and accountability when an AI system gets something wrong are exactly the questions the Caribbean AI Risk Management Council works on for real organisations. Teaching a 14-year-old to ask "what happens if this information is wrong or dishonest?" before shipping a feature is the habit that separates a useful tool from a dangerous one, not a nice extra.

Why Build It Before the Storm

The worst possible time to design a relief tool is in the middle of the emergency it is meant to serve. Adrenaline is high, information is scarce, and every shortcut taken under pressure becomes a permanent flaw. Building and stress-testing these ideas now, before hurricane season peaks, means the reasoning, the data handling and the safety rules are worked out while there is time to get them right.

There is a second reason, and it is the one that matters most for the young people involved. A generation that grows up treating AI as something that happens to them, decided elsewhere and delivered finished, is a generation that stays a customer. A generation that points machine learning and generative AI at a hurricane, the single most defining threat their region faces, and builds a real answer, is a generation that becomes a builder. Across five countries, that is the shift this challenge is designed to produce, one mixed-age team at a time, starting with an 8-year-old who already knows a good app should ask for less, not more.