Learners such as humans and intelligent agents, often require support to perform complex tasks in real world environments. Inaccessible states and the complexity of tasks add to this requirement. Large state and action spaces associated with the environments also contribute to this need for support. In this paper, we present a design of a learning agent that learns from the environment in order to support another learner. The supportive agent maps the state and action spaces associated with an environment onto state and action spaces that allow the supported learner to learn effectively. This paper presents a machine learning algorithm for the supportive agent that uses a concatenation of two reinforcement learning algorithms: a variation of a q_learning algorithm and a variation of a td (lambda) algorithm. We implemented this supportive learner for the task of training the human learners to perform prediction tasks. It is being tested on physical environments such as electrical systems and mechatronics.