Data Scientist
Current- Implemented TabNet Deep Learning model within Spark-based Fraud Engine to proactively identify and mitigate fraudulent activities in financial transactions, resulting in a $150,000 reduction in fraud losses yearly
- Developed a custom loss function that resulted in 10% fraud capture.
- Utilized Horovod for distributed training of Deep Learning models. This resulted in a 60X reduction in training time
- Migrated Spark-based application to Kubernetes, this improved scalability, reliability, and overall system efficiency.
- Deployed JupyterHub on Kubernetes, resulting in enhanced scalability automatic load balancing, and increased productivity.
Nov 2021 - Present
