Bullying isn’t what it used to be – especially considering that these days, most bullying acts occur online. Over the past two years, the data science team at L1ght set out to crack the code of cyberbullying. We will share our process of defining cyberbullying and the methodology of turning to neural networks with new and powerful machine learning models. We will explain how image recognition, speech processing, and natural language processing (NLP) taught us plenty about this behavioral phenomenon. Finally, we’ll discuss how we cracked cyberbullying using transformers, how we dealt with social networks, and share some tips and tricks.