This is an excellent book on interpretable machine learning by @ChristophMolnar [link] A project idea: read the book and start a newsletter of derived insights from as many publically available datasets as you can get your hands on.