A serious lung infection, pneumonia remains a leading global cause of death, especially among vulnerable groups like the elderly and children. Effective treatment relies on early, accurate diagnosis, often involving chest X-rays to detect pneumonia. This study examines pneumonia detection from chest X-ray images using deep learning models, specifically EfficientNetV2 variants (EfficientNetV2S, V2M, and V2L). Transfer learning was used to train and evaluate the models, focusing on balancing accuracy, inference speed, and model size. Data augmentation techniques were used to improve generalizability, while TensorFlow Lite was used for efficient deployment on mobile devices. The novelty lies in the integration of EfficientNetV2 with TensorFlow Lite to optimize performance specifically for mobile and resource-constrained environments. This unique approach balances model size, inference time, and accuracy, demonstrating significant improvements in computational efficiency for mobile applications. Results indicate that EfficientNetV2S, V2M, and V2L all achieve high accuracy: EfficientNetV2S with 98.72%, EfficientNetV2M with 98.38%, and EfficientNetV2L with 98.89%. EfficientNetV2S emerged as the most efficient model, featuring a model size of 77.17 MB and an inference time of 0.04 seconds. The TensorFlow Lite conversion reduced model size by approximately 67% and inference time by over 99%, making the models feasible for real-world use in resource-constrained environments. This research demonstrates the potential of combining EfficientNetV2 variants and TensorFlow Lite to create an effective, lightweight solution for pneumonia detection, particularly in remote areas, and suggests future work on advanced optimizations and direct comparisons with other methods