The Cloud Operations + Innovation Applied Artificial Intelligence (CAAI) team creates and delivers AI/ML models to improve data center operations. As engineer #3 for this team, I had the opportunity to work on greenfield projects and transform proof of concept models into enterprise ready services.• Productionized models, preprocessing steps, and data and training pipelines by collaborating with data scientists, deeply understanding their work, and refactoring. Models include text and image… Show more The Cloud Operations + Innovation Applied Artificial Intelligence (CAAI) team creates and delivers AI/ML models to improve data center operations. As engineer #3 for this team, I had the opportunity to work on greenfield projects and transform proof of concept models into enterprise ready services.• Productionized models, preprocessing steps, and data and training pipelines by collaborating with data scientists, deeply understanding their work, and refactoring. Models include text and image classification models and text ranking models for tasks like labelling data, anomaly detection, and selecting a part for replacement. Models often used NLTK and Tensorflow.• Refactored a ResNet based, Keras model for image classification that was expensive to run in Spark by migrating it to Kubernetes and configuring the environment to work with GPU. The change halved the cost and tripled the speed of running the model.• Augmented a text classification model for a time sensitive task management system with an interactive, real-time rule system which reduced the impact of data drift and the expense of retraining and deploying the model.• Migrated REST APIs and data pipelines from a single-region architecture to a multi-region, active-passive architecture and automated certificate rotation to support business continuity and disaster recovery. Standardized deployments between environments using ARM templates and Helm to keep the environments from drifting due to ad hoc deployments.• Hosted weekly devops meetings for 3+ years to review service monitors, triage issues, and address security and compliance tasks.• Created a dashboard with Azure Data Explorer that breaks down and forecasts cloud expenses. The dashboard replaced the manual lookup and estimate for monthly budget reviews and revealed the biggest movers and largest costs (e.g., managed Spark). Show less

Tasked with optimizing memory management in the Virtual Machine Monitor (VMM) for smoother VM experiences.- Implemented lockless data structure to reduce VMM page fault latency by 8x.- Automated testing and visualized latency improvements over current build.- Actively studied the source code and communicated with mentor. Identified several bugs in the process.- Updated internal wiki with my own experience to improve onboarding experience.

06/2016 – 05/2017. Designed a cycle level GPGPU simulator for the CUDA runtime API.06/2015 – 06/2016. Generated and analyzed memory traces of popular PC games. Reduced memory interference and improved fairness between CPU and GPU applications.