Build idea · Intermediate
An AI study buddy
Upload your notes, get a tutor that actually knows your material.
A study app where you drop in your own course notes, PDFs, and lecture slides, and an AI quizzes you on them, explains what you got wrong, and grounds every answer in your actual material instead of hallucinating from the open web. The core feature is retrieval-augmented generation over your uploads: your notes get chunked and embedded into a vector store, and at quiz time the app pulls the relevant passages so the questions and explanations stay true to your syllabus. It tracks which topics you keep missing so it can drill you harder where it counts.
What you'll learn
- →How RAG actually works end to end: chunk documents, embed them, store the vectors in pgvector, and retrieve the right passages at query time instead of stuffing everything into the prompt
- →Forcing an LLM to return reliable structured JSON (quiz questions, options, correct answer, explanation) with tool/structured outputs so you never regex prose
- →Streaming model responses token by token so explanations feel instant instead of a spinner
- →Scoping every row to the signed-in user so one student's notes and progress never leak into another's
Take it further
- +Add SM-2 style spaced repetition so the app resurfaces cards you keep failing, with a daily 'here's your review set' email via Resend
- +Voice mode: answer out loud with speech-to-text and have the buddy grade your spoken explanation, which is closer to how you'd get tested
- +Shareable study decks so a classmate can import your embedded notes and quiz off the same material