George Brown Email & Phone Number

Built personalized product discovery matrix-based affinity model that drove +50% add-to-cart rate on front page Migrated all daily model training and eval from on-premise server to serverless architecture with AWS SageMaker and Lambda Built Product2Vec recommendation models on product listing pages that drove +80% CTR

Seattle, Washington, United States

Built user embedding model: loaded and analyzed user purchase datain Jupyter Notebook, transformed data with PySpark, compared modelarchitectures based on problem statement, trained and tested model.Refactored multiple deep learning models making efficient use ofNumpy and PyTorch functions to reduce training time.Worked on automated hyperparameter.

New York, New York, United States

Designed and built a command-line tool from scratch to seamlesslyload patient health CSVs into Postgres database with minimal humanintervention and automated type and foreign key inference.Responsible for code quality, unit and integration tests, andinterfacing with stakeholders during the design process to guidedesign with their feedback.

Greater Pittsburgh Area

Teaching Assistant for Introduction to Machine Learning course with 150 students. I have office hours and lead a 150-student weekly recitation.In the past, I have been a teaching assistant for the Principles of Imperative Computation course and the Artificial Intelligence course, so I have a wide variety of teaching experience.

Greater Seattle Area

Designed, built, and tested microservice that tracked the cloud migration status of 400+ Groupon services. For each service, tracked stats including dependency versions, build status, docker image creation, and kubernetes deployment. The service itself ran in Kubernetes with a DaaS database. Held full responsibility of ownership and maintenance of the.

Huntsville, Alabama Area

Worked with the machine learning team to build a database of over 300 million documents for natural language processing tasks including similarity searching and visualization. Used SQL and ElasticSearch efficiently to build an ingestor that could add 800 documents per second. Trained Gensim Word2Vec model to create 200 dimensional Numpy vector for every.