Developed and integrated dynamic basis functions into the NBEATS model within the Nixtla/neuralforecast repository, enabling the model to adapt more flexibly to diverse time series data patterns. Focused on architectural updates that support multiple basis types, including Legendre, polynomial, and changepoint bases, the work enhanced the model’s adaptability and reduced the need for manual tuning. Leveraged Python, PyTorch, and NumPy to design, implement, and validate these changes, ensuring robust integration with existing forecasting pipelines. This feature improved forecast accuracy and flexibility, directly supporting inventory optimization and demand planning use cases without introducing new bugs during the development period.