During October 2024, contributed to the IBM/vllm repository by implementing a bad words sampling parameter within the text generation pipeline. This feature allows configurable filtering of undesirable terms during content generation, supporting stronger moderation controls and improved policy compliance for user-facing applications. The work involved frontend integration and parameterized content filtering, ensuring that moderation settings could be easily adjusted as needed. Leveraging skills in Python, machine learning, and natural language processing, the developer focused on enhancing content safety and governance. The solution was delivered with attention to code review standards and deployment readiness, emphasizing traceability and maintainability throughout the process.