This repository provides a TensorFlow implementation of QANet, a neural network architecture for machine reading comprehension introduced at ICLR 2018. The model processes questions and passages to extract answers, using convolutional encoders with self-attention rather than recurrent layers for faster inference. Training uses the Stanford Question Answering Dataset with GloVe word embeddings, and achieves EM/F1 scores around 70.8/80.1 after 60k training steps.