The educational landscape is experiencing a surging demand for Automated Essay Grading (AEG), prompting the need for innovative solutions. This paper introduces a cutting-edge methodology that harnesses the power of Bidirectional Encoder Representation from Transformers (BERT) to embed and score essays in the scientific AEG domain. Tackling challenges such as Out-of-Vocabulary (OOV), BERT's contextual embedding proves instrumental. The study meticulously evaluates a hybrid architecture on a prototype incorporating non-English essay answers, establishing a benchmark against state-of-the-art studies. Beyond the expeditious grading of essays, particularly in scientific realms, this paper makes a substantial contribution to the ever-evolving field of educational technology. The AEG task revolves around the automation of essay response grading, where input data encompasses essay answers, and output data comprises assigned scores. The adopted mathematical model seamlessly integrates BERT for contextual embedding and subsequent scoring. The evaluation uncovers compelling results, underscoring the effectiveness of the proposed BERT-based model. The model's architecture, characterized by bidirectional layers and a dense output, encompasses a notable 2,243,401 parameters. Significantly, the Kappa Score achieved by the model impressively stands at 0.9725, highlighting its superiority over existing methodologies.