Worked on the nvidia-holoscan/holohub repository to address PyTorch 3.5 compatibility issues in the Object Detection Model. Developed a wrapper class in Python to manage device placement and tensor permutation, ensuring that model inputs conformed to Holoscan requirements. Scripted the model with this wrapper to validate input formats and resolve dimension mismatches that previously caused deployment errors. Focused on configuration management and model deployment, the changes stabilized runtime behavior and reduced debugging overhead. By improving input handling and reliability, the work enhanced maintainability and paved the way for broader PyTorch version support within the Holoscan integration workflow.