From Research to App: What Our Students Build With AI Coding Tools

This season, our students built many apps across health, the environment, agriculture, sports, and education. Many build on research projects students have already done, with help from AI coding tools. Here is a description of the range of possibilities and how we think about using the tools well.

AI is now part of how students build

AI coding tools such as Codex, GitHub Copilot, Cursor, Replit, and Lovable are now a normal part of how software gets built, and students are no exception. Many student competitions now allow them, provided the student discloses how they were used, understands the technology behind the app, and makes meaningful technical contributions of their own.

So the question isn't whether students use AI to build. It's how, and what they can still explain and defend afterward. The range below shows what's possible when the idea and the understanding are the student's.

The range

Health and medicine

- A screening questionnaire paired with a chat feature that answers questions about a mental health condition using evidence from published research, built on retrieval-augmented generation

- A tool that predicts whether a brain-stimulation treatment for drug-resistant epilepsy is likely to produce a favorable short-term response, built on a student's research with brain-signal data

- A research tool that estimates whether a tumor is likely to have a high or low mutational burden from gene mutation data, with confidence and explanations

Environment, agriculture, and climate

- A web app that predicts how toxic a chemical compound is to bees

- A tool that predicts dissolved organic carbon in rivers

- A hurricane-damage prediction app that reads storm conditions from a simulated data source

- An agriculture pest detector built on an object-detection model

- A classifier for pests and diseases in grape crops

Sports and wellness

- A running-injury risk tool that gives personalized scores and explains the factors behind them

- A mobile app that builds weekly training plans for student athletes and adapts them to daily check-ins on sleep, soreness, and energy

- An app that selects music from a person's heart-rate and stress signals, positioned as wellness and not as medical care

Education

- A reading assessment tool for children that compares spoken words with the text and flags words that were correct, mispronounced, missed, or added

What they have in common

• They start with a real problem, from bees to rivers to a young reader's mistakes.

• Many sit on research the student already did. The model, the data, and the lab notebook come first, and the app makes the work usable by someone else. Research and building feed each other, as we describe in our post on student entrepreneurship.

• Students position health tools accurately, as educational or research tools, or as wellness aids, never as diagnoses.

• We push every app to show how its model works, what it can't do, and how it was checked.

The apps are at different stages. Some are working products, some are prototypes, and some are still being shaped. That's normal for a season's work.

What the tools change, and what we add

• Design thinking becomes critical. AI coding tools let students build sophisticated apps and iterate quickly. When a new version takes hours instead of weeks, the hard part is no longer writing the code. It's deciding what the app should do, who it's for, and what makes it good. That judgment is the student's, and it matters more than ever.

• We help students shape the vision. Many students don't yet know what's possible. Our mentors guide them on what an app could do and what it would take, so their ideas can grow beyond what they'd have imagined alone. The students then use AI tools to turn that vision into a working app.

  
  

• Real apps need real features.** An app that people actually use needs more than a model and a screen. It needs things like security, a database, and user accounts. We show students how these work, so their apps can become real and be used by real people.

How students use the tools

Using an AI coding tool is a skill, and so is knowing where it stops. Our approach follows the rules many competitions now set:

• The student owns the idea, the design, and the understanding. The tool speeds up the building. It doesn't make the decisions.

• AI supports specific parts of the project. It doesn't do all of the technical development, and the student makes significant contributions of their own.

• Keep a build log, and disclose what AI did, wherever the work is submitted and in conversation.

• f you can't explain it, don't ship it. Judges and reviewers may ask to see the code, and the student should be able to walk through it. A clear repository helps, as we describe our post on GitHub repos.

Key things to know.

AI makes building faster. It doesn't make the judgment about what to build, how to test it, or what claims are fair. The students whose apps stand out are the ones who understand them. This skill - being able to combine judgement and taste, with powerful tools, will carry far beyond a single project or competition. As students go to college and beyond, this skill is what we see helping them succeed in an AI-infused world.

If your student has an idea

Our Mentors in Residence help students turn research into something people can use, and use AI tools as a coach and not a ghostwriter. If you'd like to talk, we'd be glad to.

Book a free consultation here.