A defensible machine learning technique has been previously proposed which combines the understandability and logical reasoning capabilities of a rule-fact expert system with the machine learning capabilities of a neural network. This algorithm, called the gradient descent trained expert system, used a backpropagation training process and a forward-chaining-based execution/decision making process. Problematically, for some applications, both the training and execution processes are comparatively heavyweight. This paper proposes a lightweight implementation of the gradient descent trained expert system algorithm.