Lead Software Engineer
Itasca, Il, Us
Real Time Machine Learning Fraud Decisioning Platform Real-time, highly scalable, low latency solution processing 100+ million transactions per dayComplete rebuild of existing service to leverage Spark clusterLeverages historical and profile matches across dozens of fields while providing low-latency responsesAutomated extracts for Decision Science team; Quick to deploy new modelsUtilizes Machine Learning frameworks (XGBoost, Keras/Tensorflow, others)Derived Data and Service for ML ModelsReal-time service providing derived data for transactionsRetrieves hundreds of fields across billions of records and returns a response in ~4-6 msReal-time ingestion for all transaction dataSpark computations that recomputes profile, historical, fraud-to-gross, and ratios model variablesDynamic config; Recompute nightly and on demand for entire historyRequired extensive optimizations due to high volume and fluidity of data and small size of clusterBig Data Platform SupportDeveloped POCs for adding ML frameworks to existing productsXGBoost, Keras/Tensorflow, deeplearning4j, In-house gradient boosting model, log regressionsDeveloped POCs for modeling workflows leveraging SparkInstrumental in transforming the way the Decision Science team leverages dataProvided support for any critical issuesAutomated Machine Learning Model Generation for Fraud DetectionGenerated multiple model types: rule-based genetic algorithm, neural networks, svm, log. regressionDecreased run-time from several months to a couple of days per modelMachine Learning models provided a 20-50% reduction in review ratesEmployee Recognition Award for work on the automated machine learning tool