Understanding the decision-making prowess of any agent is of paramount importance,as it increases trust and guarantees safety. Having a structured and transparent representation of a policy helps us understand, evaluate, and modify it if necessary. Behavior trees due to their inherent reactivity, modularity, and transparent hierarchical representation are an ideal candidate to represent such a control policy. Reinforcement learning, on the other hand, after the advent of deep learning has shown its dominance in several domains, but struggles in deployment in high-stakes domains due to their opacity. In this study, we focus on building a knowledge transfer framework where post-hoc knowledge of trained reinforcement learning agents is captured through imitation learning, then utilized to form a compact behavior tree. Our primary focus is on retaining maximum performance while improving the interpretability of the behavior tree. By doing so, we automate the construction of a behavior tree and offer an alternative transparent architecture for policy representation.