Developed support for the SoulX-Singer singing voice synthesis and conversion model within the vllm-project/vllm-omni repository, focusing on seamless model integration and deployment. The work involved creating deployment configurations and comprehensive end-to-end testing scripts using Python, with an emphasis on audio processing and deep learning techniques. By establishing reusable scaffolding for deployment and testing, the project now enables faster onboarding of future models and improves the reliability of continuous integration pipelines. No major bugs were reported during this period, reflecting careful implementation and thorough testing. The contribution strengthened model deployment workflows and enhanced the product’s overall integration capabilities.