<p>Improving English writing ability relies on timely, multidimensional, and effective feedback. However, traditional manual correction is inefficient, and existing automated tools mainly focus on surface-level grammar checking, making it challenging to meet in-depth learning needs. Therefore, this study designs and evaluates a Transformer-based, multidimensional, real-time feedback system for English writing, called TransWrite Mentor. The system integrates a grammar correction module with a fine-tuned T5 model, a vocabulary enhancement module that fuses contextualized word vectors, and a discourse coherence analysis module based on Sentence-BERT, providing users with collaborative, interpretable, real-time tutoring across four dimensions: grammar, vocabulary, syntax, and discourse. To verify the system’s effectiveness, this study conducted comprehensive training and testing on a multi-domain corpus containing 155,000 sentences, including 40,000 sentences specifically from English learners. The experimental results show that the system’s core grammar correction module achieved an F0.5 score of 62.1 on the CoNLL-2014 test set. A further user study invited 300 subjects to complete writing tasks. The results show that the system’s vocabulary suggestions achieved an acceptance rate of 85.2%, significantly improving the text’s lexical richness. While the experimental group utilized the TWM system, the control group performed the same writing tasks using a traditional rule-based grammar feedback tool. The feedback on discourse coherence increased the logical coherence score of the experimental group’s compositions by about 18.7% compared with the baseline system. The average score on the four-dimensional user satisfaction survey was 4.25. This study confirms the feasibility and effectiveness of the Transformer-based multi-module integrated system for providing in-depth writing feedback and offers a new solution for intelligent writing instruction.</p>