I won't give details about the question as I respect the confidentiality of the interview. However, to give a general feeling, I think it doesn't hurt to mention the following. For example, code a class that implements a very popular ML algorithm. Even if the algorithm is very simple there are lots of possible improvements and generalisations, how to make it robust, efficient etc. Same thing for a class storing common data formats: dataframe, time-series, etc... how would you efficiently code access methods and/or storing according to the features of these data types?
