Ask HN: Knowledge gap between machine learning research and a functioning model?

I have been asked to implement an unkown machine learning algorithm that has been researched and proven by a team from a top University.

The goal is to build a functioning machine learning model and a series of software products around it.

For anyone with experience in this type of academic research or the software implementation of it, how much actual knowledge of machine learning should I have before attempting this?

Should I expect the research algorithms to work with minimal changes?

Or should I expect to fully rewrite the model using the research as a rough guide instead?