The paper reports on the implementation of an NLP pipeline for Bulgarian, developed within the spaCy framework and based on BulTreeBank as main source for training and test data. This new end-to-end pipeline aims to ensure easier technical maintenance and synchronization between modules, superior processing speeds -for use in real applications, and greater flexibility of adaptation. We discuss the challenges encountered in the implementation process and the solutions adopted, including the architecture itself, as well as its quantitative and qualitative evaluation.