Worked on the RoBorregos/home2 repository to enhance the GPSR node by removing blocking YOLO service calls and refactoring pose detection loading. Leveraged Python and computer vision techniques to implement a non-blocking architecture, reducing latency and increasing throughput for real-time robotics perception. Integrated TensorRT-accelerated model loading, which improved both startup and runtime performance while decoupling model management from core logic. This approach enabled more flexible inference paths and simplified future model upgrades. The work focused on maintainability and performance, ensuring the GPSR node could operate smoothly in demanding robotics environments and adapt quickly to evolving computer vision requirements.