Jan 2022 – May 2022 · 4 mos

• Developed an orchestration and transformation engine for a unified data platform. Built architecture for end-users to define custom pipelines and automate the process of refining, transforming, and aggregating data sets.

Jun 2021 – Aug 2021 · 2 mos

• Developed a unified data platform for a single source of truth to be leveraged across the enterprise in the form of a data lake. Enhanced ingestion capabilities by adding support for various source types (Sybase, Db2, Aurora). Improved dataflow by processing normalized data into Kafka structured streams piped to S3.

Jun 2020 – Aug 2020 · 2 mos

• Created Confluent Kafka connectors to route data in/out various platforms within Google Cloud Platform. Refined architecture to validate and standardize dataflow among different input formats.

Jan 2020 – Dec 2021 · 1 yr 11 mos

• Created a pop-lyric generation system by adapting the T5 transformed model for text-to-text transfer learning trained on a corpus of pop-lyrics from several decades. The architecture performs favorably in language metrics (BLEU) compared to well-known hip-hop lyric generation methods and is shown to be indistinguishable from human-generated lyrics by surveys. Conference publication to HAI 2021.
• Created a Multiclass Bayesian Classifier to attribute song lyrics to artists and genres under the supervision of Dr. Gil Weinberg and Richard Savery. Classifier reached 70% accuracy on test data among 4 different genres and 8 different artists with 1000+ songs pulled from the Genius API.

Jan 2020 – Dec 2021 · 1 yr 11 mos · Atlanta, Georgia, United States

• Developed a subversive artificial intelligence agent for countering visual algorithmic surveillance to protect personal information and vulnerable populations. Implemented universal image-agnostic perturbations for well-known deep neural nets and evaluated efficacy with human surveys.
• Created a stegonagraphical browser extension to preserve privacy of end users when sharing images on social media applications. Implemented Faugère's F5 algorithm to obfuscate user image within a stock image while resisting visual and statistical analysis for the Mozilla Firefox browser.

May 2019 – Aug 2019 · 3 mos

• Created and Dockerized a Python data analytics service to analyze hourly network connectivity and dataflow in/out of IoT Gateways in conjunction with a Flask web application to visualize service output.