Development of a multi-agent chatbot for user query resolution for UTAR
Universities generate vast amount of information daily, including programme details, course structures, schedules, policies, and procedures. This information is often distributed across multiple sources such as university websites, portals, and PDF documents, making it difficult for students,...
Saved in:
| Main Author: | |
|---|---|
| Format: | Final Year Project / Dissertation / Thesis |
| Published: |
2025
|
| Subjects: | |
| Online Access: | http://eprints.utar.edu.my/7104/1/fyp_CS_2025_HTY.pdf http://eprints.utar.edu.my/7104/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Universities generate vast amount of information daily, including programme details, course
structures, schedules, policies, and procedures. This information is often distributed across
multiple sources such as university websites, portals, and PDF documents, making it difficult
for students, staff, prospective applicants, and parents to quickly access accurate and up-to
date details. At Universiti Tunku Abdul Rahman (UTAR), this challenge highlights the need
for a unified and intelligent information access system. To address this, the project proposes
and develops a multi-agent Retrieval-Augmented Generation (RAG) chatbot designed
specifically for UTAR. The chatbot architecture employs specialized agents namely
Admissions Agent, Finance Agent, and Examinations Agent each connected to its own vector
database containing structured knowledge extracted from official university sources. Data
ingestion is automated through a web scraping and PDF download module that handles
inconsistencies such as broken SSL certificates on UTAR domains, ensuring reliable and up
to-date knowledge collection. The system integrates OpenAI API service as the base large
language model (LLM), with LangChain for orchestration, Chroma as the vector database,
Flask for backend development, and React for the frontend user interface deployed on Firebase.
The backend is deployed on Render to support scalability, concurrency, and real-time
availability. Evaluation was carried out using technical performance testing alongside user
experience testing through a structured Google Form survey. The results show that the chatbot
delivers accurate and contextually relevant answers within an acceptable response time, while
user feedback indicates strong satisfaction with ease of use, usefulness, and willingness to reuse
the system. Open-ended responses also highlighted areas for improvement, such as expanding
departmental coverage and tighter integration into UTAR’s official website. By enabling 24/7
access to official university knowledge sources, the chatbot improves information accessibility
and user satisfaction, demonstrating the feasibility of applying multi-agent RAG architectures
in the higher education context. Future enhancements will focus on integration with UTAR’s
official platforms, and extending the system to additional departments and more. |
|---|