I've done something similar at work, primarily to generate random data that'll pass our validation checks (real cities, valid SSN, etc).

It seems strange to me that the email address and usernames in the samples are based on a person's name, but they're different from the user's actual name.

I chose to just take the name, and transform it for the email address. I think some consistency will make it easier to test that the right user data shows up in the right places while testing.

I've used Faker and like it. For mocking big datasets I prefer http://www.mockaroo.com/ served by https://github.com/typicode/json-server