This project, carried out independently during my internship at UiT The Arctic University of Norway, had a clear goal: to reduce the administrative workload caused by the many emails and calls received by faculty, especially during the start of the academic year.

Prospective and current students frequently asked the same questions: program content, career opportunities, life in Narvik, housing, how the university works… This high volume took up a lot of faculty time, even though it was not their main responsibility.

Project Objective

Develop an intelligent chatbot able to automatically answer these questions reliably and in context, while providing a simple and natural experience for students.

My Role in the Project

I worked alone on this project, with a progress meeting every two weeks with my supervisor. This included:

• analysis of administrative and academic needs,
• design of the knowledge base,
• full chatbot development,
• integration of language models,
• creation of evaluation and test scenarios,
• continuous improvement based on user feedback.

User-Centered Approach

Much of the chatbot's improvement came from real-world feedback. I met with:

• students (Norwegian and international),
• faculty members,
• administrative coordinators.

They tested the chatbot, asked real questions, shared their frustrations, and I iterated to refine the quality of the answers.

Technologies and Architecture

The chatbot is based on a lightweight RAG (Retrieval-Augmented Generation) architecture tailored to the academic context:

• internal document base (courses, FAQ, local information),
• semantic search engine,
• GPT model for controlled generation,
• automated tests via AnyLogic to evaluate usage scenarios.

Project Impact

The chatbot achieved a 95% satisfaction rate, validated by both faculty and students. It correctly answers the vast majority of questions without human intervention.

The tool enables:

• a significant reduction in administrative requests,
• better availability of faculty,
• clear, centralized, and 24/7 accessible information,
• a notable improvement in the experience of new students.

IEEM 2024 Publication

This work led to a publication at the international conference IEEM 2024 (IEEE International Conference on Industrial Engineering and Engineering Management).

The paper presents the architecture, experimental results, and the evaluation methodology used.

Skills Developed

• AI & NLP (RAG, controlled generation)
• Python
• User needs analysis
• Model evaluation via AnyLogic
• Chatbot design
• Autonomous work in an international context

External Links

• Project GitHub
• Chatbot Report
• AnyLogic Simulation