408 AI Tutor
An intelligent question brushing + wrong question analysis + knowledge blind area management tool for the 408 unified examination of computer postgraduate entrance examination.
##Function Overview **Import question bank** -Upload the PDF of 408 question bank such as Wang Dao, automatically parse multiple-choice questions and comprehensive questions, identify subjects, chapters, and knowledge point tags and put them into storage. Supports direct extraction of text PDF and visual recognition of scanning PDF (VLM downgrade), and supports PUA garbled code detection and automatic repair. **ˇ Brush the questions practice ** -Brush the questions randomly or in order according to the subject/chapter, correct them instantly and display the correct answers and analysis. Support batch problem brushing mode and progress tracking. ** Answer statistics ** -Track the changing trend of correct rates in each subject, intelligently analyze weak knowledge points, and count the mastery according to the dimension of knowledge points. ** AI teaching assistant ** -Intelligent question and answer based on the DeepSeek model, which can provide in-depth explanations and answer questions on specific topics or knowledge points. ** Wrong question set ** -Automatic collection of incorrectly answered questions, support filtering by subject/chapter/status, redoing questions one by one or in batches, status marking (mastered/to be consolidated/not redone), manual addition and removal. ** Knowledge blind spots analysis ** - AI automatically analyzes knowledge blind spots and common misunderstandings in wrong questions, and generates correction suggestions and memory methods. ##Common Questions **Q: Does the AI teaching assistant function need to be networked? ** A: Yes. AI teaching assistants and knowledge blind spots analysis rely on the DeepSeek API and require effective API keys and network connections. Question brushing and question bank management functions can be used offline. **Q: What PDF formats are supported? ** A: Two types of PDFs are supported: ①** Text-based PDF**(such as Wangdao Electronic Edition title book)-Text is extracted directly through PyMuPDF, which is fast and does not require a visual model;②*** Scanning/picture-based PDF** -The system automatically detects and downgrades to the VLM visual model (Tongyi Thousand Questions VL), which is rendered page by page and then identifies the title through OCR.`VISION_API_KEY` needs to be configured. You can also check "Mandatory VLM" when uploading to skip text extraction. **Q: Why do I need two API keys? ** A: This project uses a dual model architecture: DeepSeek is responsible for reasoning (AI teaching assistant dialogue, answer analysis), and Tongyi Thousand Questions VL is responsible for vision (scanning PDF recognition, image OCR). DeepSeek's API does not support image input, so a separate visual model is required. If you only use text PDFs, just add DeepSeek. **Q: Where are the database files? ** A: Default to `data/questions.` db, SQLite single file database. **Q: How to import in batches? ** A: Click 'Import Question Bank' and select 'Import from Specified Directory'. The path is mounted to the Lazy Cat network disk by default, and'/app/home'needs to be brought by default. If the pdf file is in the 'Download' folder, fill in the path "/app/home/download", see the attached figure.
