I designed an internal platform that turns expert human judgment into training data for Reddit's safety models, built to be configured by any team, not just ours.