So first, the critical feedback - here I got click-baited into thinking the entire article would be about the PhD argument itself, which you do allude to, but I think there is a lot to expand on in that front.

As for the good stuff, your suggestions mostly overlap mine - define your OWN project (I'm less a fan of Kaggle) and start from scratch - get the data, build the models, design a pipeline. The more of the entire picture you can build yourself, the better. Too many people can fit a model these days.

Also, love the other comment on here, further cementing the "whatever I did is best for DS" bias. There are many paths to data science, mostly because data science means so many things.

P.S. - the mentor advice is sooo good.