Increase in the amount of data provides a huge scope for data analysts to operate and leverage information from them. Problems arise when the data varies in formats and their storage mechanisms become heterogeneous. Hence integration of data and its conversion to a common structure becomes mandatory. This paper presents a mechanism that operates on heterogeneous data sources, identifies conflicts and resolves them using duplicate elimination and ranking techniques. Further, a feedback mechanism is incorporated into the architecture, using which reinforcement learning is imposed on the architecture. This makes the proposed framework a machine learning architecture that is flexible and adapts according to the dynamic environment, which has become a de facto in the current scenario.