Timely and constructive feedback is essential for supporting student learning, yet providing it at scale remains a challenge for educators. Large Language Models (LLMs) offer new opportunities to automate parts of this process, but their effectiveness depends heavily on how decision logic is formulated and maintained. When such logic is embedded directly in prompts, it can be difficult for instructors to update or adapt it over time. This thesis presents a framework that integrates Decision Model and Notation (DMN) with LLM prompting to make feedback generation more modular, transparent, and easy to refine. The approach decomposes complex evaluation rules into smaller, structured decision steps, which are then used to guide the LLM’s reasoning.