The increased use of Artificial Intelligence, and in particular Machine Learning (ML) raises the need for widespread AI literacy, in three particular areas related to ML; understanding how ML works, the process behind creating ML models, and the ability to reflect on its personal and societal implications. Existing ML learning tools focus primarily on the first two areas, and to a lesser degree the third. In order to address this, we designed VotestratesML; a tool allowing K-12 students to build models and make predictions based on real world voting data. Based on in-situ deployments of VotestratesML, we reflect on opportunities and challenges for engaging K-12 students in understanding and reflecting on ML. We find that the design of VotestratesML supports students' engagement in all three areas of ML, through grounding ML in a known subject area and allowing for collaboration and competition.