VW Credit, Inc. (VCI), a financial services arm of Volkswagen Group of America, provides competitive financial services to its dealers and 1.2 million customers. The credit arm wanted to provide more timely and customized products to its customer base. VCI was working with disparate on-premises data-science platforms. Achieving its goals would require consolidated data that provide timely insights to improve price-prediction models and MLOps efficiencies to improve the time to market. Better insights and decision-making are essential for near real-time loan approvals, adaptive price prediction models, etc. These improvements would help VCI offer its customers the right mix of services exactly when needed.
Cloud migration was required to allow integration of all platforms, provide a unified view of data and provide the ability to see market changes in real-time. VCI collaborated with Wipro and the AWS team for the cloud migration journey. Wipro proposed a modular phased approach with the creation of a detailed architecture design for data model migration. Wipro leveraged the Amazon SageMaker ecosystem for the cloud migration journey and other required features like security (encryption keys, VPC), automation, notification and alerts, integrated and automated logging, monitoring services and future model scalability. The complete solution leveraged a wide range of AWS services:
SageMaker notebooks help VCI data scientists quickly build new machine learning models – from developing, building, training and deploying. Reusable frameworks expedite feature engineering, model training and model deployment.
SageMaker Model Monitoring service automated model monitoring – if there is data drift, the accuracy of models could impact the price prediction leading to potential revenue losses. Automated monitoring helps VCI capture all inference requests, compare them against baseline on a defined schedule (hourly, daily, etc.) and send timely alerts.
AWS Step Functions Data Science SDK framework, created by Wipro, helps VCI data scientists quickly create ML lifecycle workflows, effectively orchestrate all model lifecycle steps, ensures conditional checks and approvals for model deployment, and sends notifications for any failures. New machine learning models are published seamlessly to give VCI the agility to respond to changes in the market.
Amazon Event Bridge and SNS allow DevOps teams to configure and standardize type and text content for all notifications and alerts. Wipro created event-based triggers to invoke the state of machine model training/retraining, model scoring and model monitoring. SNS notification service sends email/SMS notifications for any SDK pipeline status change and success or failure events to promote faster resolution.
Amazon Cloud Formation Templates automated the deployment process with no manual interventions for code migration to the higher environment. Reusable cloud formation templates enable automated deployments, minimizing errors from incorrect code promotion. Features include version control with automated CI/CD pipelines resulting in higher code consistency across multiple environments.
AWS Lambda provides serverless computing that listens to the event notifications and executes code in response to the events. In addition, it automatically manages compute resources to drive efficiency.
AWS S3 stores objects with customer-managed encryption keys and Amazon Cloud Watch is logging and monitoring resources to ensure that data and models are secure.
For VW Credit, Inc., the benefits of the solution are far-reaching. VCI has reusable templates for model pre-processing, training, deployment templates, etc. The templates reduce the effort required for onboarding new use cases and ML models. The new architecture is scalable and can grow with future needs. The data science team has streamlined workflows to structure their code and promote collaboration and automated pipelines are improving model training and deployment. Batch-based and real-time Inference pipelines help accelerate the model deployment process, and integration with applications is helping improve model adoption. Continuous model monitoring and drift notifications help the team take quicker proactive action on data anomalies or model performance degradation.
While all these new capabilities are driving the desired efficiency and performance of the data science and DevOps teams, the features have given VCI the ability to provide new, timely and custom financial products to their customers. According to Chirag Patel, Software Engineer Technical Lead and Jaymin Mehta, Senior Manager, Big Data and Analytics, “With the help of Wipro and the AWS team, we were able to consolidate our data science platforms and automate the MLOps pipeline with proper monitoring. This will help us to achieve elasticity and scalability for our data science platform.”