This document presents a motor imagery-based brain-computer interface speller for mobile devices. It introduces a motor imagery structure model consisting of pre-processing, feature extraction, dimensionality reduction, and classification blocks. It develops an autoencoder-based dimensionality reduction method and compares it to PCA. It also develops a Hex-O-Spell mobile application using motor imagery to spell words. Results show the autoencoder approach achieves better performance than PCA. Testing on three subjects demonstrates the utility of the Hex-O-Spell mobile application. Future work involves enhancing the methods and application.