In the third installment of a series on deep learning for mobile devices, the article details the process of creating an iPhone app capable of recognizing handwritten numbers using a deep learning network. It starts with training a model on the MNIST dataset, which consists of 28x28 pixel images of handwritten digits, using convolutional neural networks (CNNs) built with Keras. The model is then converted to Apple's CoreML format for deployment on an iPhone. The author explains how to set up the app in XCode, integrate the CoreML model, and utilize the iPhone's camera to capture images for real-time number recognition. The article emphasizes the importance of preprocessing data, model training, and conversion to the CoreML format, while also highlighting potential improvements such as data augmentation and adding a "background" class to enhance model performance. It concludes by suggesting potential applications of computer vision in mobile apps and encourages readers to experiment with the provided code.