We illustrate how to use and adapt generic data science packages for visualizing and analyzing conversational data sets. Our approach combines three levels of visualisation and analysis : at corpus level we present an interactive visualisation of the diversity of conversations, at conversation level, the temporal dynamics of relevant variables are presented as dialogic time series; Finally at deeper linguistic level we deal with distributions, clustering techniques and graph visualisation to strengthen insights into language structures. We argue that only such a multi-scale approach makes it possible to grasp conversational data sets without overlooking potential pitfalls.