Last-mile delivery (LMD), the transportation of goods from a distribution hub to the final destination, is the most critical and costly segment of the logistics chain and various strategies have been developed to address its challenges. One class of approaches focuses on optimizing delivery routes for vehicle fleets, a problem known as the Vehicle Routing Problem (VRP). Extensive research has been conducted on novel methodologies and algorithms to solve VRP and its numerous variants. However, these theoretical advances fail to produce the expected results in real-world scenarios due to factors such as inaccurate travel data estimates caused by unpredictable events that may occur during the deliveries. Moreover, driver expertise is typically not considered in route assignment. This paper describes SmartDelivery, a novel platform for LMD optimization. We propose a Machine Learning (ML)-based Algorithm Selection (AS) approach to identify the best heuristic or metaheuristic from a selected portfolio depending on the characteristics of CVRP (Capacitated Vehicle Routing Problem) instances. Furthermore, we describe a method for collecting and processing real-time data to improve travel time estimates between destinations. Lastly, we define a “sixth sense” parameter quantifying drivers’ familiarity with destinations to improve route assignment. The platform is not yet fully implemented. Future research will focus on its practical deployment and validation.