I have found that in any situation that goes beyond one off research or experimentation into something that needs to be orchestrated oe deployed in production,

I have found that in any situation that goes beyond one off research or experimentation into something that needs to be orchestrated oe deployed in production, you need to move to a language with a lot of bells and whistles libraries which are usually better accessed proceedurally rather than functionally. This is because at that stage, your users are no longer people who understand your code base or are even know programming necessarily. You can do algorithm development in MATLAB or R or Julia or whatever and figure out ways to either translate the logic to Python/C++ later or setup some kind of IPC for production software engineers to call (platform/orchestration stuff is usually Python or Java)... ...OR you can just develop your algorithms in Python and keep code comprehension and platform integration much simpler for everyone