SmartCourse: A Contextual AI-Powered Course Advising System for Undergraduates

1 st Yixuan Mi*, 2 nd Yiduo Yu, 3 rd Yiyi Zhao · 2025

Read SmartCourse: A Contextual AI-Powered Course Advising System for Undergraduates by 1 st Yixuan Mi*, 2 nd Yiduo Yu, 3 rd Yiyi Zhao (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2507-22946.txt.

Introduction

Yixuan Mi, Yiduo Yu, and Yiyi Zhao present a course advising system, in a 2025 preprint, that gives undergraduates advice shaped by their own records. Standard advising tools list courses but ignore the student's history. SmartCourse, built for computer science majors, reads the transcript and the degree plan together. A locally run language model turns that context into personal suggestions. The system also handles accounts, enrollment, grades, and four-year plans, through both a command line and a web screen.

The team scored the advice on twenty-five sample questions. They invented measures for plan fit, personal fit, and recall. With full context, the suggestions were far more relevant than with context stripped out. Retakes and missing requirements were spotted when the record was visible. The authors conclude that transcript-aware advising is not a luxury but the core of useful guidance. Privacy stays local since the model runs on site.

Catalog record

CREATOR1 st Yixuan Mi*, 2 nd Yiduo Yu, 3 rd Yiyi Zhao
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE23 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2507-22946

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

SmartCourse: A Contextual AI-Powered Course Advising System for Undergraduates Thanks: * Corresponding Author 1 st Yixuan Mi*, 2 nd Yiduo Yu, 3 rd Yiyi Zhao Abstract We present SmartCourse, an integrated course management and AI-driven advising system for undergraduate students (specifically tailored to the Computer Science (CPS) major). SmartCourse addresses the limitations of traditional advising tools by integrating transcript and plan information for student-specific context. The system combines a command-line interface (CLI) and a Gradio web GUI for instructors and students, manages user accounts, course enrollment, grading, and four-year degree plans, and integrates a locally hosted large language model (via Ollama) for personalized course recommendations. It leverages transcript and major plan to offer contextual advice (e.g., prioritizing requirements or retakes). We evaluated the system on 25 representative advising queries and introduced custom metrics—PlanScore, PersonalScore, Lift, and Recall—to assess recommendation quality across different context conditions. Experiments show that using full context yields substantially more relevant recommendations than context-omitted modes, confirming the necessity of transcript and plan information for personalized academic advising. SmartCourse thus demonstrates how transcript-aware AI can enhance academic planning. The source code is publicly available at this GitHub repository . I Introduction Traditional academic advising tools often provide generic guidance, lacking integration with student-specific data such as transcripts and degree plans [ 1 ] . Existing advising solutions tend to be generic and fail to incorporate detailed student context (such as completed courses or degree requirements) [ 2 , 3 ] . In particular, there is a lack of advising systems that directly integrate a student’s own transcript and major plan into the recommendation process. More recently, advances in AI and LLMs suggest new possibilities for personalized advising [ 4 ] . Chatbot-based advisors can offer tailored course selection advice and career guidance [ 5 ] . Our system supports three types of users—students, instructors, and administrators—each interacting through role-specific interfaces. Motivated by these trends, SmartCourse addresses these limitations by embedding personalized advising directly into the academic workflow. We investigate how transcript and degree plan impact course recommendation relevance. To this end, we design controlled experiments that systematically omit these inputs and analyze the resulting drop in recommendation quality. This paper makes the following contributions: • Contextual Advising Architecture : We develop SmartCourse, a modular platform that tightly couples academic operations with AI-driven advising. SmartCourse combines core academic services with AI advising in a unified interface. • AI-Powered Recommendation Engine : SmartCourse integrates a locally hosted LLM (via Ollama) to generate course suggestions. The LLM generates suggestions based on contextual prompts. Automated email notifications can be sent when new recommendations or updates are available. • Experimental Evaluation with Context Ablation : We formulate a set of 25 realistic advising queries (e.g. elective choices for AI specialization, courses for a cyber security graduate track, GPA-improvement strategies) and evaluate SmartCourse under four context conditions (full context, no transcript, no plan, question-only). We introduce relevance metrics – PlanScore, PersonalScore, Lift, and Recall – to quantify how well the recommendations align with the student’s outstanding degree requirements, as well as latency in seconds. • Results and Insights : Incorporating transcript and plan context leads to consistently higher recommendation quality, while omitting them harms performance. These findings highlight the importance of contextual information in personalized advising, showing clear improvement over generic recommendations (see Table I for a summary of evaluation scores). The remainder of the paper reviews related work (Section II ), describes the system design (Sections III – IV ), presents the experimental setup and results (Sections V – VI ), and discusses key findings, limitations, and future directions (Section VII ). II Related Work Course recommendation in education has been studied through collaborative filtering, matrix factorization, and content-based methods [ 6 , 7 , 8 ] . These techniques often rely on preference data or grade history but typically ignore structured curricular requirements [ 9 ] . Meanwhile, academic dashboards and degree audit tools help students monitor progress and register for courses, but rarely offer personalized or adaptive suggestions. SmartCourse addresses this gap by aligning recommendations with formal degree plans and transcript records. More recently, LLMs have been explored for educational use cases [ 10 ] , such as chatbot-based tutoring, FAQ answering, or summarizing curriculum content [ 11 ] . While LLMs can offer natural-language interactions, they often lack access to structured academic records, leading to generic or hallucinated advice [ 12 ] . SmartCourse builds upon these strands by embedding a locally hosted LLM into an end-to-end course management system [ 13 , 14 ] . Unlike prior work, it fuses transcript, degree plan, and user queries into contextual prompts, enabling personalized and curriculum-aligned advising within a unified platform. III System Overview III-A Architecture Overview SmartCourse is organized into several interacting components. It maintains student and instructor accounts (with roles and majors), supports course management (registration, prerequisites, and grades), and enforces a four-year degree plan for the specific major. A central AI Recommendation Engine uses a local LLM to answer student questions based on their academic record. The system provides both a text-based CLI and a user-friendly Gradio web interface for interaction. Fig. 1 illustrates the overall SmartCourse architecture (data inputs, modules, and UI layers). Fig. 1: SmartCourse system architecture. Users interact via CLI (administrator) or Gradio GUI (instructor & student), managed by the CourseManager layer. Data is stored in flat-file repositories, with an LLM providing recommendations and a mailer handling notifications. III-B Core Components SmartCourse comprises the following building blocks: • User Accounts : Credentials, user roles (student, instructor or administrator), and declared majors are stored in a secure account store with hashed passwords. • Course Enrollment & Grading : A course-catalog repository lists all available courses. Students enroll through either interface; instructors record grades, which are written to an enrollment ledger that tracks course codes and grades per student. • Degree Plan Management : For each major (e.g. CPS) a standard four-year plan repository defines required courses by year. The system compares this plan against the transcript to monitor progress and identify outstanding requirements. • AI Recommendation Engine : When a student poses a question, SmartCourse constructs a prompt that fuses the question text, current transcript, and four-year plan context. The prompt—structured as transcript, plan, and question—is sent to a locally hosted LLM (specifically llama3.1:8b via the Ollama runtime). The returned answer is parsed and filtered into valid course recommendations (Fig. 2 ). Fig. 2: AI-generated advising response based on integrated transcript and degree plan context. This response was rendered as a schematic interface for clarity and does not depict a live system screen. Fig. 3: Administrator interface for managing accounts, courses, and AI model configurations via CLI. Simulated terminal view created for illustrative purposes; not a direct screenshot. Fig. 4: Instructor interface for assigning grades and viewing student course records. Drawn representation of the interface layout; the actual GUI may differ visually. Fig. 5: Student interface for registering and dropping courses, viewing progress, and requesting AI-based advising. This interface depiction is illustrative and not captured from a running system. • Interfaces : We illustrate these interfaces below to provide a concrete view of the system’s user experience. Administrators operate through a dedicated CLI to maintain system-level configurations.

Read the complete text