Open Source in Finance Forum New York 2022

Real-time analytics could help financial institutions create enhanced customer experience, combat fraud, and manage risk better. This talk uses credit card fraud detection to demonstrate how to combine the power of open source and closed source to build and train the AI model for fraud detection on the Cloud and deploy it for real-time fraud detection on the mainframe at scale. Credit card transactions are typically processed on a mainframe. Training the model is more convenient in a Cloud setting where the on-premise transaction data and other relevant third party data are brought into a secure environment compliant with financial industry regulations. This talk shows how to build and train deep learning models using TensorFlow on the Cloud and persist the trained model in an open standard format (ONNX). Finally the trained model is deployed on the IBM mainframe where Telum AI Accelerator is used for real-time in-transaction inference at scale. The pattern described in this talk could be applied to other use cases in financial industry and beyond. Open source and closed source software can work in harmony to solve many real-world problems in a hybrid multi-cloud environment.