Delivering personalised and timely feedback is crucial for helping students address gaps in their understanding. However, the increasing demands of large class sizes make this task particularly challenging for CS1 educators, especially for casual teaching assistants who lack formal training and experience. Existing feedback training methods are often inconsistent and ineffective, leaving educators unprepared to handle diverse student needs.

To address this, we designed an adaptive fading procedure based on Cognitive Load Theory (CLT) to support educators in delivering high-quality, personalised feedback. This pedagogical technique adjusts instructional support to varying learning needs and is integrated into textit{FeedbackPulse}, an automated tool that evaluates feedback in real-time and provides tailored guidance for improvement. This paper outlines our approach to designing scalable, evidence-based feedback instruction using Generative AI and large language models (LLMs) to overcome challenges with feedback quality in CS1 education.