The use of machine learning in driver assistance systems allows to significantly enhance their functionalities. In particular, it allows to personalize systems by evaluating the driver’s past behavior. Such personalization is especially relevant for recommendations in maneuvers where the specific maneuver embodiment strongly depends on the driver’s momentary driving style and attention. Led by this idea, PRORETA 4 developed a prototypical City Assistant System, which gives the driver a personalized recommendation in urban scenarios. To adapt the recommendations and warnings appropriately, the system incorporates the learned momentary driving style and the driver’s gaze behavior. In this work, we describe the main functional blocks of the system, present our solutions to major implementation challenges and also discuss the safety of the used learning algorithm.