Medical AI is rapidly becoming a reality. Building, training, and deploying AI within healthcare requires tackling unique challenges. Steps include acquiring and curating clinically relevant datasets, performing high-quality annotations, training on huge multislice datasets, and deploying into a complex HIPAA-centric IT infrastructure. We'll describe our approach for a cloud-based pipeline from data ingestion to trained model deployment. We'll describe a deployment strategy where trained deep learning models are run entirely in a browser within the bounds of a hospital's network. This can be done using our open-source library, Keras.js (supporting Keras with Tensorflow/Theano backends). Inference times on real-world networks are made practical through GPU acceleration using WebGL.