Worked on the unslothai/gpt-oss repository to deliver offline model serving capabilities for VLLM, enabling local execution of language models without requiring internet connectivity. Focused on decoupling model serving from web access, the solution reduced latency and eliminated reliance on external APIs, making it suitable for edge and offline scenarios. The implementation included clear setup instructions and predefined conversation templates to streamline deployment and usage. Leveraged Python and machine learning deployment skills to integrate the offline serving feature, providing practical guidance for users. The work addressed the need for flexible, self-contained model deployment in environments with limited or no network access.