Document Type

Thesis

Degree Name

Master of Applied Computing

Department

Physics and Computer Science

Program Name/Specialization

Applied Computing

Faculty/School

Faculty of Science

First Advisor

Abdul-Rahman Mwlood-Yunis

Advisor Role

Thesis Advisor

Second Advisor

Xu (Sunny) Wang

Advisor Role

Committee Member

Third Advisor

Ilias Kotsireas

Advisor Role

Committee Member

Abstract

Academic advising is fundamental to student success, yet traditional practices and models struggle with scalability, consistency, and the dynamic nature of institutional in- formation. While Artificial Intelligence (AI) and Large Language Models (LLMs) offer promising solutions, their deployment in this critical domain has been hindered by issues such as factual inaccuracies (hallucinations) and the static nature of knowledge bases. This thesis addresses these challenges by developing and validating an intelligent and dynamic student advising system powered by Retrieval-Augmented Generation (RAG). The research first conducts a comprehensive study, demonstrating that a RAG-based chatbot significantly outperforms fine-tuned LLMs in terms of factual accuracy, response relevance, and scalability for academic advising tasks. RAG is a preferred method in this application because it ensures AI responses are grounded in verifiable, current institutional data. Building upon this foundation, the thesis presents a structured methodology for the construction of a high-quality, university-specific knowledge base from diverse unstructured data sources. To overcome the limitations of standard RAG implementations, this work introduces an advanced RAG architecture. This includes a two-stage retrieval process initiated by a fast, embedding-based router that calculates the semantic similarity between the user’s query and the available documents. This method intelligently identifies the most relevant source document to narrow the search space, thereby significantly enhancing retrieval precision without the computational cost of an extra LLM call. Complementing this, a dynamic knowledge management framework is developed, en- abling the addition and deletion of documents from the knowledge base without costly re-indexing. Collectively, these contributions yield a robust, accurate, and highly main- tainable AI-driven academic advising system. By mitigating hallucinations, improving retrieval efficiency, and ensuring dynamic knowledge base updates, this framework signif- icantly enhances student support, reduces advisor workload, and sets a new standard for intelligent information delivery in educational environments.

Convocation Year

2026

Convocation Season

Spring

Download

Available for download on Wednesday, October 06, 2027

Share

COinS