Generating Student Curriculum Plans with Local Large Language Models

Christopher Eng

Co-Presenters: Abijith Manikandan, Luis Velazquez Rodriguez, Joseph Tomasello

College: The Dorothy and George Hennings College of Science, Mathematics and Technology

Major: Information Technology

Faculty Research Mentor: Daehan Kwak

Abstract:

The increasing adoption of Large Language Models (LLMs) has prompted widespread exploration into their potential applications across numerous sectors of the labor market. Businesses and academic institutions are progressively examining how integrating AI-driven conversational models into their workflows can address specific organizational needs, optimize resource allocation, and significantly reduce the time and effort required for various tasks. A particularly promising area of investigation involves utilizing the Natural Language Processing (NLP) capabilities inherent in contemporary LLMs to automate labor-intensive administrative tasks. This project specifically aims to evaluate the efficacy of currently available LLMs in generating personalized academic curriculum plans for students enrolled in college degree programs. Central to this assessment was determining the capacity of LLMs, varying by origin, technical specifications, and deployment environments, to accurately interpret and execute a set of interrelated instructions. These instructions guided the placement of multiple courses into an organized, semester-based schedule format. Precise prompt engineering was essential for enhancing the models' accuracy and ensuring effective generation of individualized curriculum schedules. The outputs generated by these models were subsequently reviewed for adherence to specific constraints, such as course prerequisites and semester-specific course availability. The primary test configuration consisted of a structured, four-semester schedule, each semester comprising four courses, with only one correct arrangement meeting all prerequisite and availability requirements.

Previous
Previous

Evaluating Automatic Speech Recognition​ Models: How Well Do They Handle Accents?​

Next
Next

Information Gain-Based MCDM Approach for Urban Critical Freight Corridor Identification using Spatial-Temporal Data Analysis