10/24/2024
Hackathon
An AI-powered platform that reshapes learning paths in real time.

Why we built it

At ShellHacks 2024, my teammates Colin, Trevor, and I wanted to test how far AI could push education. Most learning apps move linearly — ask a question, check the answer, move on. We wanted something different: a system that would adapt dynamically to each learner, reshaping both the content and the path in real time.

AI Learning Platform Appimage/png • 994KB • 5088×3306

The Solution

  • Custom AI SDK: Before “AI SDKs” were common, we engineered one from scratch with modular event hooks (setOnAnswerReceiveData, setOnRegisterTopic, etc.), so any interface (CLI, web, game) could plug into it.
  • Dynamic questioning: The system didn’t just generate multiple-choice questions; it tracked performance, adapted difficulty, and shifted topics based on user mastery.
  • Knowledge Graph: Every topic was registered as a node with prerequisite/child relationships, forming a real-time skill tree that mapped what you knew and what you should learn next.
  • Explanations First: When you got an answer wrong, the SDK forced the model to explain why — ensuring comprehension, not just correction.
  • CLI + Web Frontend: I built the frontend in Next.js (rebuilt it three times during the hackathon to keep pace), while Colin crafted the backend SDK logic. We even wired a CLI so we could test and demo without a UI.

Impact

  • Delivered a working adaptive tutor prototype in under 36 hours.
  • Proved that LLMs could scaffold learning by building difficulty ramps and topic graphs on the fly.
  • Created a system that went beyond “quiz app” — it was closer to a personalized tutor with memory, reasoning, and adaptability.

Reflection

This project was one of the most energizing weekends of my life. It was my first deep dive into building with AI as a force multiplier — not a replacement for creativity, but something that let us ship faster and think bigger.

We were building the early shape of what many “AI tutors” today are still trying to become: a system that doesn’t just give answers, but understands where you are, why you’re stuck, and how to move you forward. This project relit my love for programming.

© 2026 Kevin Willoughby. All rights reserved.