Developed a TTML-compatible BERT model for the tenstorrent/tt-metal repository, focusing on robust architecture design and seamless integration of Safetensors-based weight loading. The work progressed from establishing a compilable baseline to refining input sequence handling, ensuring that example usage accurately reflected the model’s configuration. Using C++ and YAML, the implementation emphasized model serialization and training workflows, reducing the risk of misconfiguration and enabling portable weight management. This foundation supports efficient benchmarking and prepares the model for production deployment. The approach demonstrated depth in deep learning and NLP, with careful attention to stability, validation readiness, and alignment with TTML inference requirements.