The github api return the memory footprint of each programming language per repository e.g: C++: 3000bytes, Ruby: 2000+ bytes. It's just about retrieving your list of public repos and going though all of them. Solution is not perfect, I assume 1 byte = 1 char, 25 char is 1 line, plus frameworks like ruby mess up the count, but it's something :D
Does it go by how GitHub tags each repo or by an independent look at the code in each repo? Maybe by file extensions?