Quizgen AI: An Intelligent Multiple-Choice Question Generation System Using Large Language Models for Educational Assessment

Abstract

Automated question generation (AQG) from educational text is an active research domain that reduces the cognitive burden on educators while ensuring consistent, scalable assessment creation. Existing rule-based and early neural approaches are limited in semantic depth and contextual coherence. This paper presents QuizGen AI, a web-based intelligent system that leverages the LLaMA 3.2 large language model (LLM) via the Ollama inference framework to automatically generate multiple-choice questions (MCQs) with explanations from arbitrary educational text. The system integrates a Flask-based REST API backend, SQLite persistent storage, and a structured JSON prompt engineering pipeline to ensure output quality across three difficulty levels. PDF export is supported via ReportLab. Evaluations demonstrate that the system generates syntactically correct, contextually relevant MCQs with an average generation latency of under 8 seconds for five questions on commodity hardware, achieving a structural validity rate of 96.4% across 500 test runs. The system maintains persistent quiz history and supports authenticated multi-user access.