How AI Built 200 Quiz Questions in 12 Minutes

The Document Parsing Pipeline

This is what happens when you upload a PDF to Auto Mode:

1. Text Extraction: Using pdf-lib for PDFs, mammoth for Word docs. We strip headers/footers and normalize formatting.
2. Chunking: Split text into 500-word segments. Each segment gets processed independently for parallel execution.
3. Embedding: Convert text to vector embeddings using text-embedding-3-small. This captures semantic meaning.
4. Question Generation: GPT-4-turbo with a carefully crafted prompt. The prompt asks for 3-5 questions per chunk.
5. Validation: Each question is checked for clarity, answer accuracy, and difficulty level distribution.

// AI prompt for question generation
const prompt = `
You are an educational expert creating quiz questions from technical content.
Create 4-5 multiple choice questions that test comprehension, not memorization.
Format: Question | Option A | Option B | Option C | Option D | Correct Answer
Content: ${chunk}
`;

const completion = await openai.createCompletion({
  model: "gpt-4-turbo",
  prompt: prompt,
  max_tokens: 500,
  temperature: 0.7
});