Talking about experiences with a person or a machine has been shown to improve emotional well-being and mitigate loneliness. Our work is inspired by prior work that has shown how poems can be developed into meaningful interactions and how poetic conversations can support emotional well-being. In this paper we report development and evaluation of a mobile & social voice-based chatbot named Kavi that talks in poetic style. Kavi’s back-bone is a generative AI model and the front-end is a web-interface that runs on most mobile or desktop devices. We trained Kavi with poem datasets instead of conventional conversational datasets. This provides Kavi with a unique poetic language style and a different dialogue style in comparison to many existing chatbots. With Kavi, we investigate 1) if Kavi can have meaningful conversations with people and 2) if varying the relevancy of the poem responses affects the interactions.