During July 2025, Sofiia Chorna focused on improving device management in the lab-cosmo/pet-mad repository, specifically addressing initialization reliability for the MADExplorer component. She centralized device handling logic to ensure the pet model is consistently moved to the correct target device, reducing device-mismatch errors and supporting stable multi-device experiments. This work, implemented in Python using PyTorch and machine learning best practices, enhanced both the correctness and maintainability of the codebase. By providing a well-documented, targeted fix, Sofiia laid the groundwork for future performance optimizations and improved traceability, demonstrating depth in addressing cross-device challenges within complex ML systems.