The goal was to design an Oracle database and test it against a realistic amount of data, not three rows typed in by hand.
The schema
We followed Merise: the conceptual model first, then the logical one. Before writing the creation scripts, we checked functional dependencies and normalised the tables up to 3NF.
The data
To fill the database I wrote Python scripts using Faker. The tricky part was
keeping foreign keys valid and values consistent across tables. Then we wrote
analysis queries (joins, correlated subqueries, GROUP BY and HAVING) and
processed the results with Pandas.