I am currently working on managing APIs and streams that interact with customers on both mobile and desktop views of Capital One’s auto-loan financing services. I gained experience keeping these APIs and streams secure through vulnerability remediation, and also using acceptance test driven development and modular program structure.

I worked on building end-to-end automation solutions for manual processes involving reporting legal and compliance results to ensure Capital One’s Refinance organization is consistently well-managed, leveraging only internal tools. This included interactive reports with live data, data pipelines and data transformations, and automated reporting of results using AWS services. I also worked on Tableau and Quicksite dashboards and SQL query optimizations.

I worked on performing image auditing using classification, segmentation, and labeling. I also worked on delivering an in-house auto-labeling and auditing infrastructure for image auditing using LabelStudio, which saved the company $120,000 per year in auditing fees. I also gained experience with enterprise machine learning data management in the cloud.

I also worked on the in-production Mask-RCNN model used for RGB image segmentation in robot vision. I gained experience building and training machine learning models from image datasets. My focus was speeding up inference time and increasing prediction accuracy by successfully completing a model conversion to TensorRT. The converted model had an inference time speedup of 6x and an inference accuracy increase from 82% to 94%.

I worked on the industry leading Palladium Z1 emulator and future products. My focus was on metric driven memory and performance optimization for a core software module that represented the multi-user system in memory and was responsible for mapping a design to a designated set of resources. I also worked on regression testing and code coverage for developed features, and gained experience working on a multi-million line code base and test driven development. I achieved a 31% decrease in memory usage by converting to dynamic data structures and memory access controls.