Inspiration

Most study tools treat learning like a straight line. Topic 1, Topic 2, quiz, repeat. But when you're actually learning, it feels like a messy web. You get stuck because one missing idea blocks five other ideas. We wanted to build something that shows that hidden structure, not just another flashcard study app.

What it does

NEXUS turns a topic into a skill tree. You start with a concept map where nodes are concepts and edges define prerequisites. Concepts stay locked until their prerequisites are mastered. To "master" a concept, you must show that you understand the concept by explaining it in your own words (typed or voice). The app transcribes voice and then uses Gemini to grade how solid your explanation is, and then if its good you unlocks what comes next.

How we built it

  • Frontend: React + Vite, Tailwind, Three.js with react-force-graph-3d for the interactive 3D graph
  • Backend: Node.js + Express, REST API
  • AI/APIs: Google Gemini for concept tree generation and explanation grading , ElevenLabs Speech to Text for transcribing user concept explanations
  • Voice: ElevenLabs Speech to Text for transcription

Challenges we ran into

  • Getting Gemini to output the structure we needed consistently (nodes, prerequisites, clean JSON), small prompt changes made a big difference.
  • Making the 3D concept dependency graph readable and usable (it got chaotic).
  • Keeping frontend and backend in sync.
  • Handling API rate limits without letting the system "pass" someone by accident.

Accomplishments that we're proud of

  • A working end-to-end loop: generate a skill tree, pick a node, explain it, get graded, unlock progress.
  • Voice input that actually turns into a transcript and having it graded by Gemini.
  • A graph UI that feels like a game skill tree, not a boring chart.

What we learned

  • Learning really fits a graph model better than a checklist, prerequisites are the real blockers.
  • AI features need guardrails. Things like validation, retries, and safe failure behavior matter as much as the model call itself.
  • Integration work is most of the struggle in a hackathon. Wiring, edge cases, and making everything work together reliably.

What's next for NEXUS

  • Add accounts to track progress per user in MongoDB.
  • Further optimize skill tree and grading prompt and tracking user data with suggestions tailored for them.
  • Add class or team mode so groups can share skill trees and track progress together.

Built With

Share this project:

Updates