Designing and maintaining distributed and componentized tools for data scientists to build ML models at scale using Python, Dask, Kubeflow, and XGBoost.

• Achieved a 40% reduction in time on a 50% increase in data in a model fit pipeline by optimizing Dask configurations and usage

• Standardized repo and engineering practices to provide consistency between repos

• Incorporated (and documented incorporation strategies of) better environment manager into components while abiding by proprietary and strict CI/CD processes, accelerating local development by 100x in some cases

Designing and maintaining distributed and componentized tools for data scientists to build ML models at scale using Python, Dask, Kubeflow, and XGBoost.

• Reduced KFP (Kubeflow Pipelines) pipeline.yaml size by 13.7% for Capital One's largest Kubeflow pipeline, allowing engineers flexibility to develop new features without hitting the pipeline file limit size enforced by KFP.

• Created a reusable recursive feature eliminator Kubeflow component, unifying customers and streamlining support under a single component

• Parallelized a data migration workflow to increase transfer speed from months to days, enabling the decommissioning of an old expensive Kubernetes cluster, ultimately saving $50,000+

Worked on the Data Lineage team using Scala, ZIO, and AWS services to manage how Capital One handles their data lineage.

• Implemented a "Write Once, Read Many" pattern with Scala 3 and ZIO 2 to remove a bottleneck in the lineage system

• Reduced monthly DynamoDB storage costs by ~55% through serverless deletion process while mitigating excessive strain on the rest of the system from CDC

Implemented the collection and appropriate querying of data for the game Rubber Bandits. The results from this process helped and continue to guide the decisions of the company as to what features to develop next and how successful these subsequent features are. Additionally, I started development on a new minigame within the Crime Spree game mode that has been published to production.