In order to accomplish automatic English proficiency scoring, deep learning techniques and technologies are successfully implemented in this study. The NAIST Lang-8 Learner Corpora and the C4 200 M dataset are two essential datasets that are at the heart of our research. We carefully implement advanced deep learning models, such as RNNs and attention mechanisms, to evaluate their influence on the assessment of English competency. With the surprising accuracy of nearly 91%, this creativity greatly improved the output of the scoring system. Moreover, the attention model is better in the F1 score compared to its non-attention equivalent, indicating that it is more capable in this context. These results focused the capability of deep learning models to provide accurate, comprehensive, and customized feedback in educational settings, possibly changing the face of language assessment and learning.