Worked on enhancing the jeejeelee/vllm repository by implementing support for the bad_words parameter in the Completions API, bringing its functionality in line with the chat endpoint. The development involved updating the CompletionRequest model using Pydantic to accept and process the new parameter, ensuring seamless integration with existing API structures. Python was used extensively to handle the logic and data validation, while comprehensive unit tests were added to verify correct mapping to SamplingParams and robust behavior across different scenarios. This work focused on expanding API capabilities and maintaining code reliability through thorough testing and adherence to established development practices.