Computer Vision and Data Analysis:

Conducted a code review of a machine learning app to measure the efficacy of new Move Over Slow Down laws. I utilized TensorFlow and Python scripts to improve the data preprocessing and data pipeline for third-party use. I also helped coordinate work between team members in Academica, state Traffic Management Centers, and the DoT.

Data analysis and visualization:

Analyzed and built a web dashboard using data collected from the plea bargaining process in Berkshire, MA. As plea agreements are under-studied but are 90-95% of all case dispositions, this work is vital to shedding light on a huge obfuscated section of our legal system. I worked to gain data-driven insight and visualize disparities by defendant demographics, reveal charge and sentencing patterns, and track judge severity.

Currently working to create a dashboard framework in javascript that can be used by any county to generate their own plea-tracking dashboard!

NLP and webscraping:

Built Google Trends and LinkedIn webscrapers, collecting time-series keyword data and business demographic data. This data is being used to find correlations between the size of Paycheck Protection Program loans given before and after the death of George Floyd by race of business owners and the financial institution issuing the loan.

Computer vision and synthetic data generation:

Trained GAN model to generate synthetic, labeled satellite imagery, blended style transfer. This was used to augment a wind turbine image training dataset collected from Google Earth. This method of artificially generating new training imagery increased our object detection model's average precision by 22%. The work of detecting and mapping energy infrastructure provides vital information for energy planning.