TestCraft: AI-Powered Personalized Question Paper and Assessment Generation System

Preparing an assessment requires teachers to select syllabus topics, balance marks and cognitive demand, check questions, and plan follow-up practice. This paper presents TestCraft, a web-based assessment system that brings these tasks together while retaining teacher review and approval. PDF and DOCX syllabi are converted into structured units, topics, subtopics, and concepts; teachers then define blueprints controlling question types, counts, marks, diflculty, and Bloom levels. The implementation combines Python and Django with OpenAI GPT-4o-mini, Sentence Transformers, cosine similarity, SQLite, and ReportLab. Simpler local parsing and generation fallbacks support development continuity when the external AI service is unavailable, without being treated as equivalent to the generative model. Generated questions undergo structural checks, semantic screening, and teacher approval before oflcial use. Configurable institutional formats, online assessments, deadline notifications, topic-level summaries, and rule-based personalized practice complete the workflow. The final automated suite passed 345/345 tests with zero failures and zero errors, and the Django system check reported no issues. Functional verification covered syllabus-dependent structured analysis, teacher-configured question generation, pairwise semantic screening, and configurable paper rendering. Analysis outputs vary with document content, while question composition follows the teacher-defined blueprint. These findings support the tested software functions; they do not establish question quality, complete duplicate detection, or improved learning outcomes.