During October 2024, Alvasian developed a configurable bad words sampling parameter for the IBM/vllm repository’s text generation pipeline. This feature allows users to filter out undesirable terms during content generation, directly supporting content moderation and policy compliance. Alvasian’s approach involved frontend integration and parameterized filtering, ensuring seamless interaction between user input and backend processing. The implementation, written in Python and leveraging machine learning and natural language processing techniques, focused on enhancing safety and governance for user-facing deployments. Although no bugs were fixed during this period, the work demonstrated depth in unit testing, code review readiness, and commit-driven traceability.