## Inspiration Students lose hours digging through PDFs and slides, and general chatbots often give answers that don't match what the teacher actually taught. We wanted an AI tutor that only answers from the class's own material and shows exactly where every answer came from.

What it does

Macro AI is a personalised learning platform for students and teachers.

  • AI tutor with citations: answers only from course material and links to the exact page, slide or timestamp. If the material doesn't cover a question, it says so instead of guessing.
  • Adaptive quizzes: MCQ, short answer and numerical questions whose difficulty follows each student's mastery of every topic.
  • Teacher tools: upload PDFs and notes, organise topics, build quizzes, publish courses and track class progress.
  • Student dashboard: streaks, mastery, weak topics and goals.
  • Permission-aware search and five themes with dark and light modes.

How we built it

  • Frontend: React 18, Vite, TypeScript and Tailwind, hosted on Vercel
  • Backend: Supabase (Postgres, Auth, Storage, Edge Functions)
  • AI: Claude, called from a server-side Edge Function so the API key never reaches the browser
  • Security: Row Level Security on every table; roles are verified in the database, and teachers join only with a single-use admin invitation

Mastery is tracked per topic with Bayesian Knowledge Tracing. After each answer, the chance that a student has learned topic \( k \) is updated and then moved forward by the learning rate \( P(T) \):

$$ P(L_{t+1}) = P(L_t \mid \text{answer}) + \big(1 - P(L_t \mid \text{answer})\big)\,P(T) $$

The next quiz question is picked to match that mastery, with unseen questions first.

Challenges we ran into

  • Keeping the tutor grounded so it never invents answers outside the course material
  • Secure teacher onboarding: choosing "Teacher" on a form must not grant teacher access
  • Opening private files safely with short-lived signed links that work in every browser

Accomplishments that we're proud of

  • 60 automated database security checks and 21 end-to-end browser tests
  • Every tutor answer is traceable to its source
  • Quiz answers are graded inside the database, so answer keys never reach students

What we learned

How to combine Row Level Security, serverless functions and a large language model into an app that is useful and safe for a real classroom.

What's next for Macro AI

  • Semantic (vector) search
  • Text extraction from slides, Word files and video transcripts
  • Course join codes and class invitations

Try it out

Team

Done by S Lalith Adithya and V Balamuruganand

Built With

Share this project:

Updates

Submission history