Contrary to how it is often painted, finance rarely goes above undergraduate level stats. If you have a reasonable handle on that, you are already well set on becoming a quantitative researcher. If not, doing a couple of courses will give you a good grasp on most of what you are likely to encounter. If you like math, that's 50% of the battle won, because learning about the various methodologies will be an exciting discovery and not a slog.

The well kept secret of the industry is that complicated stuff rarely works, and if it does, people often don't exactly know why. Put differently, a good algorithm has obvious and intuitive failure cases. That is, it is (relatively) easy to reason under which market conditions will it work well and when it will not. Extremely complex blackboxes are tricky because it's not obvious when and why they stop working. They are also very easy to overfit. Every single additional parameter you add to a model is an additional degree that you are likely to overfit just because of survivorship bias.

It is for this reason that the absolute vast majority of systematic trading relies more on intimate market knowledge than very secret sauce and super complex ML strategies. Sure, many funds run some token ML crap (which is often quite vanilla, like gaussian processes or decision trees) with very minor impact on PNL, but this is purely for marketing reasons.

There is a small handful of companies who have the manpower, expertise and infrastructure to run more complicated models, but even there, the fundamental ideas tend to be relatively simple, and the complexity arises from the amount of data and market idiosyncrasies.