KNOWPLAN: Knowledge-Driven AI Agents for Smart Degree Pathway Planning

Shuheng Cao, Jiaqi Wu

Keywords: Large Language Model, Artificial Intelligence, Data Mining, Recommender Systems, Course Planning

Abstract: Recent advances in large language models (LLMs) provide powerful capabilities for knowledge-driven course planning. However, building reliable, constraint- aware study planners from publicly available course webpages remains challenging due to heterogeneous data sources, complex multi-logic prerequisites, and multi-requirement constraints. To address these challenges, this paper proposes KNOWPLAN, a proactive, self-evolving multi-agent AI platform that integrates LLM-based extraction, knowledge-graph construction, and constraint-aware reasoning to generate adaptive, personalized study plans. This platform brings together scientific inquiry, technical challenges, and practical utility within a coherent and unified framework. The scientific inquiry focuses on two fundamental problems: the heterogeneity of publicly available university catalog webpages, and the limitations of traditional graph structures in handling prerequisite logic. The technical challenges include extracting structured course information from diverse catalogs, modeling prerequisite structures as hypergraphs, and extracting critical paths under multi-dependency conditions. To tackle these issues, we propose a multi-LLM-driven Agent Forest to handle webpage heterogeneity, introduce the Logic Adjacency Matrix as a novel representation of course prerequisite graphs, and develop the Multi-Dependency Critical Path Extraction algorithm to support effective course planning. These components represent the core technical highlights of this work. On the engineering side, a major contribution of this work is the design of a modular end-to-end pipeline composed of four key components: the Agent Forest, Graph-Construction Agent, Course Planning Agent, and Curriculum Alignment Agent. LLMs are integrated at various stages of this pipeline to support course information extraction, prerequisite cycle resolution, personalized course recommendation, and term-level schedule generation tailored to individual preferences and academic backgrounds. Across multiple universities, KNOWPLAN achieves 99.5\% accuracy on major requirements and 98.7\% on prerequisites. By combining graph-based reasoning with a term-level scheduler, it generates feasible and personalized study plans that respect preferences, workload limits, and policy exceptions, outperforming state-of-the-art methods.

Primary Area: applications to computer vision, audio, language, and other modalities

Submission Number: 5391