In the education sector, CRM plays a key role in managing the interactions and communication between students and college staff. This research aims to present a next-generation approach to improving student-staff relations by leveraging advanced NLP models to address the feedback given by the students more effectively. We employed several transformer models for text classification to integrate in a full-stack framework for real-time complaint classification. In this work, the comparison of the different types of BERT models for complaint classification is presented. The proposed BERT base uncased model was found to give the highest validation accuracy of 100% on both 90–10 and 80–20 train-test splits, thus making it the most effective in this classification task and in enhancing feedback management. The proposed model achieved outstanding performance in precision, recall, and F1 score across different averaging methods (Macro, Micro, and Weighted), with each metric reaching a score of 1.00. Additionally, it achieved an AUC-ROC of 1.0, a Matthews correlation coefficient of 1.0 on both the 90–10 split and the 80–20 split, a log loss of 0.086 on the 90–10 split and 0.103 on the 80–20 split, a hamming loss of 0.0, and a Cohen's kappa of 1.0 on both splits.