Yasodhara Varma Rangineeni

● Designed and managed a large-scale Machine Learning platform with petabytes of storage and over 50K cores, enabling efficient development and training of ML models in Python. Supported advanced ML tools such as XGBoost, TensorFlow, PyTorch, Scikit-learn, Pandas, Matplotlib, and standard Anaconda packages. Demonstrated expertise in distributed model training using EMR Rapids, Dask, and Ray, optimizing scalability and performance for complex machine learning workflows. ● Led the design and onboarding of on-premise Big Data and AWS Cloud platforms, ensuring seamless integration and scalability. Implemented robust governance and control processes, designed effective backup and recovery mechanisms, and introduced new tools in alignment with firm-wide compliance standards to enhance platform capabilities and operational efficiency. ● Managed the development and implementation of credit card acquisition and fraud detection models, leveraging TigerGraph (Graph Database) for advanced cluster analysis and multi-hop network exploration to identify and prevent fraudulent activity. ● Established AWS platform for machine learning (ML) model development and training, migrating ML workflows from on-premise Hadoop clusters to AWS services such as EMR, S3, EBS, EC2, and SageMaker. Transitioned the on-premise ML training environment to AWS ML Services, leveraging SageMaker Studio Notebooks and SageMaker Custom Images for efficient model development and training. Designed and implemented pipelines to seamlessly transfer ML feature data from on-premise systems to the cloud, ensuring data integrity and accessibility. ● Automated platform tools to detect unauthorized data usage, generate real-time alerts for compute and storage utilization, manage data archiving, and identify unauthorized software installations. Developed and implemented a GuardRail API to optimize EMR compute and notebook costs, achieving operational savings of $500K per month while enhancing resource efficiency and cost control. ● Evaluated vendor software products, including TigerGraph, ensuring alignment with firm-wide controls and compliance requirements. Managed the procurement process and developed deployment plans for seamless integration of new solutions. Led proof-of-concepts (POCs) for ArthurAI to assess advanced model monitoring capabilities and AWS Feature Store to establish a centralized repository for reusable feature engineering data, driving efficiency and innovation in machine learning workflows. ● Led the development and implementation of the First Payment Default (FPD) Model, designed to generate scores for ranking applications based on the risk of credit abuse fraud scams. Utilized the XGBoost Gradient Boosting algorithm to identify and capture non-linear patterns in credit abuse behavior. The model was trained on a diverse and extensive dataset, incorporating both internal attributes and external bureau data to ensure accuracy and robustness in fraud detection. ● Led the development and implementation of Check Fraud and Inclearing Models, utilizing machine learning techniques and neural networks for image recognition to effectively detect deposit fraud and clearing check fraud. These models enhanced fraud detection capabilities and ensured more secure transaction processing.