Efficient last-mile delivery in densely populated emerging economy cities faces significant challenges, including severe traffic congestion, unpredictable travel times, limited transportation infrastructure, and high customer expectations for timely deliveries. To address these complex challenges, this paper proposes an integrated framework leveraging crowd-generated data and advanced Artificial Intelligence (AI) techniques. The framework employs real-time traffic predictions based on sentiment analysis and geo-tagged data from public digital platforms, effectively capturing dynamic urban conditions often missed by traditional sensors. Using a novel combination of Graph Convolutional Networks (GCN) and Long Short-Term Memory (LSTM) neural networks, the developed model accurately forecasts traffic congestion by extracting both spatial and temporal dependencies. Subsequently, an adaptive deep reinforcement learning algorithm optimizes delivery routes and schedules in real time, adjusting dynamically to fluctuating demand and rapidly evolving traffic conditions. A comprehensive case study applied to Bogotá, Colombia—a representative emerging megacity—demonstrates the framework's effectiveness, achieving significant improvements in delivery performance and operational efficiency. The study provides valuable insights into how crowd-sourced digital data and AI can be strategically combined to revolutionize last-mile logistics in emerging economies, contributing toward more sustainable and resilient urban environments.