ExamReplica

Students waste time generating generic practice problems that don't match their professor's style, difficulty, or topic emphasis. Feedback from generic AI tools isn't tied to structured concept-level weaknesses, so weak areas don't get targeted attention.

An exam practice tool that builds a Professor profile from uploaded slides, homework, and prior exams (topic distribution + question patterns), generates distribution-aware practice sets aligned to a chosen scope (e.g. Notes 3–8), and grades submissions with concept-level error classification — then regenerates targeted practice for weak areas.

• Upload slides, homework, and prior exams
• The system builds a Professor profile (topic distribution and question patterns)
• Generate distribution-aware practice sets or full exams (MCQ, FRQ, or mixed)
• Submit answers; the system grades and classifies concept-level errors
• Receive targeted practice on weak concepts

• Structured document parsing + topic tagging
• Style-conditioned question generation
• Structured rubric-based grading
• Concept-level error classification with adaptive regeneration

• Frontend: React (TypeScript)
• Backend: Python + FastAPI, Docking for PDF parsing
• Database: PostgreSQL
• AI: Gemini API + embeddings (RAG pipeline)
• Tools: Antigravity, Cursor, Figma