From my personal experience, DIY projects are the most important cornerstone to build sufficient knowledge once you got the basics with your degree.

The problem with just being a graduate is that you have never solved a problem with your own collected dirty data. You did not need to come up with different methods to solve the issue until then. The perfect solution was already presented to you by someone and you just needed to implement it. I am not talking about following a youtube tutorial. This is also just some form of bootcamp. What I am talking about is that you have an idea that in theory you think you are able to solve, but by doing so you realize how difficult certain steps in the process are that you underestimated, like setting up the infrastructure, collecting your own data, clean it, create models that work with your data etc. If you do a couple of those in different areas, like one in nlp, one in computer vision, you should be much better prepared and "ready to work" than someone with a perfect graduate degreee that has never seen data outside his college courses.