Integrating Structured Knowledge with LLM Reasoning

KNOWPLAN addresses the limitations of standard LLMs in academic advising—specifically their tendency to hallucinate degree requirements or ignore complex, multi-year prerequisite chains. The framework functions as a knowledge-driven agent that decouples the reasoning capabilities of an LLM from the rigid, factual constraints of a university curriculum. By utilizing a Knowledge Graph (KG) to represent degree requirements, course dependencies, and institutional policies, the system ensures that any generated pathway is mathematically and logically valid before it is presented to the student.

The Agentic Workflow

The framework operates through a multi-step agentic process that balances student-specific goals with hard constraints:

  1. Knowledge Retrieval: The agent queries the Knowledge Graph to extract the current state of the student's progress and the remaining requirements for their specific degree program.
  2. Constraint-Aware Planning: Instead of relying solely on probabilistic token generation, the agent uses the KG to filter potential course sequences. This ensures that every suggested path respects prerequisites, credit limits, and semester-specific course availability.
  3. Goal Alignment: The agent incorporates student preferences (e.g., interest in specific electives, desired graduation timeline) into the planning process, allowing for personalized pathways that remain within the bounds of academic policy.

Why Knowledge-Driven Agents Outperform Standard LLMs

Traditional LLM-based assistants often fail in degree planning because they treat curriculum requirements as 'soft' information rather than 'hard' constraints. KNOWPLAN shifts this paradigm by:

  • Reducing Hallucinations: By grounding the agent in a structured KG, the system eliminates the risk of suggesting non-existent courses or ignoring critical prerequisites.
  • Handling Long-Horizon Planning: Academic planning requires looking 2–4 years into the future. The agentic approach uses the KG to maintain state consistency across multiple semesters, a task where standard LLM context windows often struggle.
  • Explainability: Because the pathway is generated based on explicit graph traversal, the system can provide clear justifications for why a specific course must be taken at a certain time, increasing student trust in the AI's recommendations.