So it has come to the point that you are advised to “find your own dataset and produce some *interesting* analyses”. It’s even better if that dataset is somewhat less popular, or, the task is more ambiguous, less clearly defined, and there are not hundreds of available notebooks on Kaggle. From my (a probably a lot of people's) experience, such datasets are as messy as they can be, with missing entries, having too many/too few observations, and all the features being in their rawest form possible.
