This summary discusses the challenges of maintaining a positive atmosphere for players while preventing toxic behavior in online gaming communities. Traditional methods such as post hoc reporting and keyword-based filtering are shown to be ineffective, leading to lower engagement and lost revenue. A scalable, real-time, AI- and machine learning-based detection system using Confluent's data streaming platform paired with the Databricks data intelligence platform is proposed to identify and respond to toxic messages without disrupting the natural flow of in-game chat. The system uses a specialized local model to quickly flag potentially problematic interactions and a deeper analysis service in Databricks for robust AI analysis and decision-making. The architecture allows for effective moderation while preserving enjoyable player interactions, maintaining a positive community, and ensuring long-term success for games.