India's rapid urbanisation has precipitated a severe air quality crisis, yet 47% of the populationresides in unmonitored "blind spots". This data deficit hinders the application of GIS and AIessential for health-supportive urban planning. Because of this spatial sparsity, traditionalinterpolation methods, such as Kriging or Inverse Distance Weighting (IDW) are not effective dueto their inability to capture the non-linear, dynamic phenomena. This study addresses the criticalchallenge of monitoring air quality in unmonitored urban sectors of Ahmedabad, India. It presentsa hybrid Spatio-Temporal Graph Neural Network (ST-GNN) framework that fuses Sentinel-5Psatellite data with urban spatial features and meteorological variables. By integrating GraphAttention Networks (GAT) for spatial dependencies with an optimized XGBoost regressor, themodel achieves high predictive precision. The results demonstrate a R2 of 0.727 for PM2.s and aR2 of 0.809 for NO2. This framework provides a scalable solution for high-resolution, citywidepollution mapping to support data-driven urban policy. It can estimate pollution levels inunmonitored "blind spots". This study describes the framework's architecture, data processing,and its validation, establishing a new benchmark for scalable, cost-effective air qualitymanagement systems. These systems are crucial for meeting the goals of the National Clean AirProgramme (NCAP).