Chest radiography is one of the most preliminary forms of radiological investigation used globally. As such, it accounts for the gateway to detecting a plethora of pulmonary anomalies, the early diagnosis of which serves as one of the primary phases of diagnosis. Contributions in this field have been plentiful with the advent of more advanced algorithms. This study proposes a novel ensemble model for Pneumonia detection that combines the predictions made by multiple miniature convolution networks that perform feature extraction from various parts of the scans. Having these localized networks allows us to obtain a highly accurate prediction and visualization of the anomalies that might be present in the scans. Additionally, since the miniature networks are only working towards finding anomalies in the relevant sections of the radiographs, it allows a huge reduction in the amount of training parameters and training time, allowing us to build complex convolution networks to perform deeper feature extraction. The proposed model acquired an accuracy of 96% on the Mendeley dataset and 96.1% on the VinDr dataset.