The existing end-to-end task oriented dialog systems that use knowledge base to generate a response, perform well only on open domain dialog system evaluation metrics, a simple analysis shows that there exists a huge scope of improvement when evaluated using task specific metrics. The failure is mostly due to the architecture's inability to handle OOV words, inability to perform simple reasoning over knowledge base results. In this talk, we present our new architecture that helps in solving these issues and attains state-of-the art results in end-to-end task oriented dialog systems.