TLDR: SmartCourse is an AI-powered academic advising system for undergraduates, specifically Computer Science majors. It integrates student transcripts and degree plans with a local large language model to provide personalized course recommendations. Experiments show that using full student context (transcript and plan) significantly improves the relevance and quality of advice compared to generic or context-omitted approaches. The system aims to offer tailored guidance, helping students navigate their academic requirements more effectively.
Navigating academic pathways can be a complex task for undergraduate students, often leading to generic advice that doesn’t fully consider their unique academic journey. Addressing this challenge, a new system called SmartCourse has been developed, offering an integrated and AI-driven approach to course advising, specifically designed for Computer Science (CPS) undergraduates.
SmartCourse stands out by moving beyond traditional advising tools. It integrates crucial student-specific information, such as their academic transcripts and four-year degree plans, to provide highly personalized recommendations. This means the system doesn’t just offer general guidance; it understands what courses a student has already taken, their grades, and what they still need to complete their degree, including any courses they might need to retake due to low grades.
The system is designed to be user-friendly, offering both a command-line interface (CLI) for administrators and a web-based graphical user interface (GUI) for instructors and students. It handles various academic operations, including managing user accounts, course enrollment, grading, and tracking progress against four-year degree plans. At its core, SmartCourse uses a locally hosted large language model (LLM) to power its personalized course recommendations.
The AI Recommendation Engine is a key component. When a student asks for advice, SmartCourse creates a detailed prompt for the LLM, combining the student’s question with their transcript and degree plan. This rich context allows the AI to generate suggestions that are not only relevant to the curriculum but also tailored to the student’s individual academic history. For instance, it can suggest electives for a specific specialization like AI or recommend courses to improve GPA.
The developers conducted experiments to evaluate the effectiveness of SmartCourse, particularly focusing on how much the context (transcript and degree plan) impacts the quality of recommendations. They tested the system with 25 different advising queries under four conditions: full context (transcript and plan), no transcript, no plan, and question-only. The results clearly showed that providing both the transcript and degree plan significantly improved the relevance and quality of the recommendations. Without this context, the AI struggled to provide useful or accurate advice, often suggesting irrelevant courses or nothing at all.
For example, in the full context mode, SmartCourse recommended an average of about 6.6 courses per query, with a high percentage of these recommendations aligning with unmet degree requirements or suggesting retakes for low-grade courses. This demonstrates the system’s ability to offer practical and actionable advice. While there were occasional instances of the LLM suggesting courses not on the official list, the system includes post-filtering to ensure recommendations are valid.
Also Read:
- DUET: An AI Tool for Diagram Feedback in Computer Science Education
- Boosting Personalized Learning: A New Framework for Smarter Question Recommendations
SmartCourse represents a significant step forward in academic advising by demonstrating how AI, combined with institutional data, can provide more effective and personalized guidance to students. While there are ongoing efforts to expand its support to more majors, incorporate user feedback, and enhance its interface, the current prototype validates the core idea of context-aware AI for academic planning. You can learn more about the technical details and findings in the full research paper available at this link.
