AutoMidterm

AutoMidterm is an easy but effective tool for automatically generating exam questions from course lecture materials to facilitate efficient exam preparation. It is designed for courses with strong alignment between lecture content and exam coverage, where lecture slides (as PDFs) are available. The system can be easily adapted to different courses by modifying the prompt template. Originally developed for the 2025 Fall Cognitive and Reasoning course at PKU.

Motivation

Preparing for midterm exams often requires synthesizing large amounts of lecture content into testable questions. AutoMidterm automates this process by leveraging large language models to generate questions and answers directly from lecture PDFs, significantly improving study efficiency. The tool is particularly effective for courses where exam questions closely follow lecture materials, especially when instructors also leverage AI for question generation.

File Organization

AutoMidterm/
├── slices/              # Place your lecture PDF files here
├── questions/           # Generated JSON question files (one per lecture)
├── questions_txt/       # Formatted text outputs
├── generate_question.py # Main script for question generation
├── generate_txt_files.py # Script for organizing questions into text format
└── prompt.txt           # Prompt template (customize for your course)

Note: Due to copyright restrictions, I cannot provide the original course materials. However, I've included generated questions in the questions/ folder for reference.

Environment Setup

Install dependencies:

pip install -r requirements.txt

Set your OpenAI API key:

export OPENAI_API_KEY="your-api-key-here"

Usage

Step 1: Prepare Lecture Materials

Place your lecture PDF files in the slices/ directory. Each PDF will be processed independently.

Step 2: Generate Questions

Run scripts/generate_question.py to generate questions from all PDFs in the slices/ directory:

python scripts/generate_question.py --num-questions <num> --model <model_type>

Key Parameters:

• --num-questions: Number of questions to generate per lecture (default: 20)
• --model: Model to use (gpt-5, gpt-4o, gpt-4-turbo, gpt-4o-mini)
• --slices-dir: Directory containing PDF files (default: slices)
• --output-dir: Output directory for JSON files (default: questions)
• --lecture: Process only a specific lecture file
• --prompt-file: Custom prompt template (default: prompt.txt)

Questions are saved as JSON files in the questions/ directory, with one file per lecture.

Step 3: Organize Questions into Text Format

Run scripts/generate_txt_files.py to organize all questions into readable text files:

python scripts/generate_txt_files.py --questions-dir <questions_dir> --output-dir <questions_txt_save_dir>

This generates three files:

• questions.txt: Questions only (numbered)
• answers.txt: Answers only (numbered)
• questions_and_answers.txt: Questions and answers paired together

Parameters:

• --questions-dir: Directory containing JSON question files (default: questions)
• --output-dir: Output directory for text files (default: questions_txt)

Customization

To adapt AutoMidterm for a different course, modify prompt.txt to reflect the course-specific requirements, question styles, and answer formats. The prompt template uses {num_questions} as a placeholder for the number of questions to generate.