Worked on the MonashDeepNeuron/Neural-Cellular-Automata repository to implement dynamic learning rate scheduling using a Turbulation Sigmoid Function, enhancing training stability and convergence reliability for deep learning experiments. Refactored the learning rate logic to leverage Python’s statistics module and ensured optimizer parameter groups were updated correctly throughout training. Added safeguards to defer learning rate adjustments during the first epoch, stabilizing model initialization and reducing the need for extensive hyperparameter tuning. Focused on delivering a robust, production-ready feature using PyTorch and Python, with an emphasis on numerical methods and machine learning engineering best practices to prepare the codebase for future development.