Educational assessment has shifted from a “one size fits all” model to learner-centered, responsive, and adaptable dynamic rubrics and feedback procedures. Dynamic rubrics boost engagement and assessment confidence by improving instructor-student communication and transparency.  Dynamic rubrics could improve feedback and assessment. Pre-processed standard dataset texts are used. Pre-processed texts are delivered to Word2Vec to extract key features and vectorize them. A fuzzy clustering model with dynamically weighted rubrics evaluates the assignment. The dynamic rubric clearly outlines the assignment's evaluation criteria, and weights assist in deciding how much each criterion contributes to the overall grade using the Foraging Phase Updated Addax Optimization (FPUAO) algorithm. Subsequently, the assessment score is obtained, and based on this score, feedback is generated using a Generative Attention Long Short-Term Memory (GenA-LSTM) model. Finally, the developed model provides the optimal responses to the students by using the dynamic rubric with a deep learning model and an enhanced optimization algorithm. Experimental results on two datasets show an average accuracy of 99.36%, representing a 2.7% improvement over K-Means Clustering.  The average Mean Absolute Error (MAE) is 31.85%, an 81% reduction compared to K-Means. Average efficiency increases by 21.30% compared to other models. Thus, the developed model is more effective and robust than the existing approaches.