Anti-Grooming Classifier & Chatbot

Online grooming is difficult to detect precisely because it mimics ordinary, trust-building conversation until it doesn't. I led development of a prototype machine-learning classifier and chatbot to identify grooming behavior in chats with children, trained on real, known grooming conversations.

The work was a collaboration between Save the Children and Omdena, whose distributed community of AI engineers built and iterated the models. The prototype was named to IRCAI's Global Top 100 projects using AI to address the UN Sustainable Development Goals.

It is an example of the harder half of child-safety technology: not just flagging known illegal content, but modeling patterns of behavior – carefully, and with a clear sense of the false-positive costs.