Worked on the AI-Hypercomputer/JetStream repository, focusing on backend development to address a critical bug in token utilities. The main contribution involved correcting the decoding method for Token IDs during text generation, ensuring that tokens were accurately interpreted throughout the pipeline. This Python-based fix improved the reliability of generated outputs and reduced the likelihood of downstream errors, providing a more stable foundation for future enhancements in token processing. The work demonstrated strong debugging and root-cause analysis skills, with careful use of version control to deliver a targeted patch. No new features were added, but overall system robustness was measurably improved.