Design and Implementation of an AI-Powered Plagiarism Detection System for UNN Postgraduate Theses

The integrity of academic research at Nigerian universities is under increasing threat from plagiarism, particularly in postgraduate thesis submissions, where manual detection methods are unreliable and resource-intensive. This study designs and implements an AI-powered plagiarism detection system for the University of Nigeria, Nsukka (UNN), capable of comparing submitted documents against a curated repository of academic publications and prior thesis submissions. The system employs natural language processing techniques, including TF-IDF vectorization and cosine similarity scoring, complemented by a fine-tuned BERT language model for semantic similarity detection. The backend was developed using Python with Flask, and the frontend built in React.js. The document corpus was indexed using Elasticsearch for efficient large-scale comparison. Development followed an iterative research prototype methodology. Evaluation was conducted using 80 postgraduate thesis excerpts, half containing known plagiarized segments at varying levels of paraphrasing. Results indicate that the system detected exact-match plagiarism with 99 percent precision and paraphrased plagiarism with 84 percent precision, outperforming a baseline TF-IDF-only model by 21 percentage points. Processing time per 10,000-word document averaged 14 seconds. The study concludes that AI-enhanced plagiarism detection substantially improves the reliability of academic integrity screening at UNN. Integration with the university's thesis submission portal and regular corpus updates are strongly recommended.