Over the past years, this technology has been praised as one of the most advanced systems for detection and prevention of fraudulent financial transactions at scale. The credit card readers that are installed in retail outlets generate an enormous, constant flow of transaction data which needs to be processed very fast in order to assess the risk, make inferences and act on events. Existing fraud solutions depend on rule systems that run off-line, thresholds checks, and review processes that require human intervention, which are slow, unscalable and ineffective against new forms of fraud, non-linear type of anomalies or sequential patterns in transections real-time flow. These are addressed with this project by developing the AI platform for automated real-time fraud detection employing Apache Kafka, Apache Spark streaming, machine learning classifiers deployed to cloud environment with analytics visualization layers allowing fast proactive event handling. It uses Apache Kafka for real-time distributed event layering where a producer sends POS transaction streams into partitioned topics under Kafka umbrella. Such broker guarantees messaging durability due to his being replicated alongside message ordering across his partitions as well as failure tolerance among broker nodes. By means of Kafka partitioning ordering parallelizing events consumption and production can be reached so vertical scalability is provided regarding high-thorough put transaction environments thus far so good replicas de relojes system reliability system wide established support level preventive action has improved.