Shoutout to the guy at the Denver library who caught my resume sorter flagging nurses as bad hires
I was testing my little AI resume sorter in a study room at the Denver Public Library last Tuesday with about 200 sample resumes, and it kept dumping every nurse into the reject pile. Turned out my training data was mostly office jobs, so the model learned that words like "clinical" and "shift" meant unqualified, and I fixed it by feeding in 50 real nursing resumes and checking the picks by hand. How do you spot that kind of quiet bias in a small model before it does real damage?