Developed a PyAV-based video backend for the jeejeelee/vllm repository, focusing on enabling concurrent video decoding to improve frame loading performance under heavy load. The work involved updating the video loading logic and integrating automated tests to ensure reliability and maintainability of the new backend. By leveraging Python and PyAV, the solution addressed throughput and latency challenges in video-heavy workloads, particularly during peak usage. The implementation emphasized backend development and video processing, with clear commit practices and collaborative sign-offs to support code quality and knowledge sharing. No major bugs were reported, reflecting a stable and well-integrated feature release.