The main problem is each subsection of a chapter in NR is really a subfield in and of itself, and to adequately describe a decent modern implementation is at least one if not multiple books.
First make sure you're familiar with background issues in numerical analysis/computation such that you're generally familiar with IEEE floating point and its foibles, numerical stability, basic error analysis etc.
Figure out which algorithm you're interested, find a couple of open source implementations (academics generally generate a vast array of these, though many may become abandonware), look at any papers they are associated with/mention and read those. Now you're ready to start writing.
Once you have a basic implementation, most fields have a set of standard problems they benchmark on. Find those and see how you compare on numerical accuracy and performance against the libraries you looked at before.
> First make sure you're familiar with background issues in numerical analysis/computation such that you're generally familiar with IEEE floating point and its foibles, numerical stability, basic error analysis etc.
You might find the book Matters Computational: Ideas, Algorithms, Source Code by Jorg Arndt (the book is freely available) and the accompanying FXT library useful - https://www.jjj.de/fxt/fxtpage.html
The Eigen library (https://eigen.tuxfamily.org) has great attention to numerical robustness and accuracy in it's implementation.
The documentation is good, but for the numerical discussions, you'd have to read the source code.
I get that for most just using Numpy, GSL or similar library would be sufficient, but what if you really want to implement them yourself.