Fellow on the code generation team working on Codex which powers GitHub Co-Pilot. Scaling laws for data quality with improvements resulting in a 2x performance boost on HumanEval

Research intern at Google Brain.Contributed key mathematical and algorithmic insights into a new self-supervised pretraining method that leverages recent advances in differentiable sorting for representation learningDefined new state of the art performance for audio and vision based tasksStreamlined and standardized several scattered experiments across notebooks and code bases. This drastically increased team productivity and we were able to launch multiple large scale experiments dailyDesigned and developed two novel metrics to measure experimental success now used by the team to communicate our findingsOrchestrated foundational experiments across thousands of GPUs and decreased data loading time from 5 minutes to 300msExplored mathematical relationship between entanglement and optimal transport distance, presented work to teamReproduced results from the field which increased our ability to iterate and improve upon existing research

Prediction intern on the Level 5 Self-Driving team. I worked on dynamical vehicle prediction using physics, machine learning, and data driven models. I improved the performance of the system 22.5%I wrote high performance, modern, c++ for this project for real time feature extraction and trained models using Python. These models were deployed to the car with Tensorflow. I also worked on an internal semantic code search tool using modern transformers, NLP, and Pytorch

ML Engineering intern on NLP / Text understanding team.Researched and implemented phishing detection tool using sophisticated feature engineering, random forest, and logistic regression techniques. Achieved ∼96% accuracy with a .005% false positive rateEngineered and developed asynchronous API using parallel processing and high performance computing techniques to achieve a 3x speed up resulting in a 63% reduction in hardware costsThe final estimated impact of this project is $300k - 500k in savings.Built question similarity tool using sentence embeddings after collecting and curating a dataset of ∼130,000 questions. Improved f1 score from .3 to ∼.7Read, implemented, and improved methods for emerging topic tracking from state of the art papers. Resulting in a functioning prototype that allowed my team to reach their stretch goals for the quarter

Designed and taught a competitive coding course to talented math majors with topics focusing on algorithm design, problem solving, data structures, and mathematical familiarity for solving programming questions. Of my students, 12 placed in the top 20 at the Fall 2017 ACM coding competition.Our team placed in the top 50 in the annual hackerrank collegiate challenge

Member of Amazon Alexa Prize Team EveDesigned and built an offensive speech filtering system using probabilistic methods, which performed ~3% better than current industry standardsResearched and designed a complex sentiment analysis tool that didn't only classify sentences as positive, negative, or neutral but also determined if the sentiment was complex.Research AssistantCollaborated on machine learning project researching low cost MRI configurationsDeveloped machine learning models to improve relatively poor quality imaging allowingfor wide spread medical imagingDeveloped parallel solutions to augment arbitrary image data sets, and simulate MRI results reducing processing time by 300%Designed and built a deep learning platform to reduce background noise for hearing aid users resulting in a system that reduced the signal to noise ratio by 197%Identified and assisted with various bug fixes to learning algorithms and probabilistic programming modelsAssisted in formulating convergence proofs for DFA like Deep Learning learning systemsMost of the work done involved Pytorch. We did all of our work on Linux systems with Docker and Cuda.

Developed super powered web solutions to significantly increase employee effectiveness, resulting in significant annual savingsCollected, cleaned, and analyzed internal and external data to pinpoint areas of business strength and weaknessOptimized daily processes, and increased page load speed by 75% (Python, MySQL)

Combined technical knowledge with business acumen to ensure apps were useful to consumers and accurate in their functionsPersonally created 160 Domo business apps that use data to tell the customer’s story (JS)Wrote killer documentation for customers

Discovered proper boundary condition equations to more accurately model pressurewaves using numerical methods, resulting in a method of approximation that was 3xfaster than previous methodsCollected data on models, and tested our results for stability and consistencyPresented research findings to experts, professors and fellow students

Developed custom web games and did user testing to help refugees learn EnglishExcelled in Data Science course work earning top marks (Python)Presented secondary research to peers and mentors

As a member of the BYU studies web team, I assisted in the maintenance of our ASP.NET systems, integration of new functionality, maintenance of our mySQL server, extensive front end design, and related administrative requirements.Led a web team of 3 in maintaining a website with thousands of unique visitors,increasing traffic and profitability by over 38%Managed a relational database, customer information, and new projects

Worked independently to develop firm's website, and sites for local companies. Assisted on the creation of electrical plans for University of Washington