Inspiration
Students spend countless hours searching through lecture notes, slides, PDFs, and online resources, yet still struggle to understand difficult concepts. We wanted to build an AI learning companion that goes beyond simple question answering and helps students truly learn by providing grounded explanations, personalized quizzes, and adaptive revision plans.
What it does
The project transforms lecture notes, textbooks, and PDFs into a personal AI tutor. Instead of only retrieving information, it identifies knowledge gaps, explains concepts with citations, generates practice questions, and continuously adapts to the student’s progress.
How we built it
We built the project using Codex and GPT-5.6 throughout development.
Core technologies include:
- GPT-5.6 for reasoning and tutoring
- Codex for implementation and rapid iteration
- FastAPI backend
- React frontend
- Retrieval-Augmented Generation (RAG)
- Vector search for document retrieval
- SQLite for local data management
Codex accelerated development by helping implement features, refactor code, generate tests, and explore alternative implementations while we focused on product design and user experience.
Challenges
One of the biggest challenges was reducing hallucinations while keeping responses useful. We solved this by grounding every answer in uploaded learning materials and encouraging the model to cite relevant sources. Another challenge was creating an adaptive tutoring experience rather than a simple chatbot, requiring careful prompt design and conversation flow.
What we learned
We learned that building an educational AI product is not only about language models, but also about retrieval quality, user experience, and learning psychology. Working closely with Codex also changed the way we develop software by making iteration much faster while keeping humans responsible for product decisions.
Future Plans
We plan to add spaced repetition, learning analytics, collaborative study spaces, and multimodal support for handwritten notes, images, and recorded lectures, making the assistant a complete AI-powered learning platform.
Log in or sign up for Devpost to join the conversation.