I was contracting at GMAC-RFC (the home mortgage arm of General Motors) when the crash started.
I was working on software models to predict what would happen with low-quality mortgages. There was certainly an effort to figure out which mortgages were likely to be bad and which weren't. I didn't do the verification, but it looked good to the people that did (in a statistical way; all we could do was figure out probabilities). Personally, I wondered why there were any mortgages with "stated income" and "stated assets", in which the mortgage company didn't verify income and assets - liar's loans, as they were called.
Another issue was that real estate prices stopped going up. The bad mortgages were often sold to naive buyers with the assurance that, if the borrowers couldn't manage to pay the mortgage, they could always sell the house for enough to pay off the mortgage plus a little. As long as this was true, nobody would really lose that much. The bank would wind up with someone paying a mortgage, and the home buyers would wind up with some of their investment back, as well as having been able to live in a house. When they peaked and then declined, it left a lot of people with a mortgage larger than the market value of the house ("underwater"), and when they couldn't pay the mortgage everybody suffered. Another thing I wondered about was why, in the model, there were parameters for how fast housing values would increase, but they didn't go negative.