Worked on the vespa-engine/sample-apps repository to enhance the reliability of the BPE tokenizer for long-text processing scenarios. Addressed a critical bug by implementing a context length guard that trims tokens when input exceeds the maximum allowed context length and ensures the end-of-text token is set correctly. This solution prevents token overflow and downstream errors, improving stability for applications handling extensive text inputs. The work involved applying expertise in Natural Language Processing, text processing, and tokenization, using Java as the primary language. The focus on correctness and boundary enforcement contributed to more robust handling of long-form content in downstream models.