gygy-open/question-bank

Question Bank is a system for building and managing a library of exam or quiz questions, built with AI baked into its core workflow rather than added as an afterthought. It is aimed at educators or organizations that need to collect, organize, and reuse test questions at scale. The standout feature is smart importing: a user can upload a Word document, a Markdown file, or even an image of a printed test, and the system uses AI to automatically pull out structured questions from it, including multiple choice, multiple answer, fill in the blank with support for more than one correct answer, true or false, and open ended questions, along with rich text and math formulas. Imported questions go through a review workflow of draft, pending review, published, and archived, with a full audit log and support for soft deletes and bulk actions. The system also organizes questions by knowledge point using a technique called RAG, retrieval augmented generation, where a vector database called ChromaDB finds candidate topics related to a question and then a batch reranking step maps those candidates back to a standard set of subject knowledge points. It supports multiple AI providers, including Google Gemini, OpenAI, and any OpenAI compatible API such as DeepSeek or a privately hosted model, and these can be swapped without restarting the service since the configuration lives in the database. The project also includes an AI chat feature that supports images, exam assembly from filtered questions into a temporary basket for export, subject and tag management, user roles and permissions, and full activity logging. The frontend is built with Nuxt and Vue, and the backend runs on FastAPI with Python. The whole system can be started with one Docker Compose command that brings up the database, vector store, backend, background worker, and frontend together. It is released under the AGPL-3.0-or-later license.