As a member of Credit Karma's Machine Learning Infrastructure team, I implemented automated hyper-parameter optimization. I performed extensive experiments to select hyper-parameters to be tuned and chose the value ranges to be explored during automated tuning.
My work involved developing new Python implementations of model training pipelines. To accomplish the automation of hyper-parameter tuning I leveraged Google Cloud Platform (GCP) services like, AI Platform, Google Cloud Storage (GCS), BigQuery (BQ), as well as Airflow, CircleCI, and scikit-learn.
My work delivered sizable performance and cost-efficiency improvements. More accurate models allow Credit Karma's members to see more precise approval odds for recommended credit cards and personal loans. These product improvements in turn deliver member engagement, customer satisfaction, and revenue growth to Credit Karma.