Created and implemented algorithm mapping difference of architectures to difference in accuracy, which projected the problem to a lower dimensional space and provided a host of benefits across a diverse set of benchmarks including a 50% improvement in sample efficiency in NASBench201 (WACV ‘25 in review)Developed k-Oneshot networks in combination with an augmented predictor to address domain specific issues in EDA that yields 40% closer to ideal performance (0.9802 ROC-AUC and query time of only 0.461 ms) (DATE ‘25 in review)Harnessed locality to shrink large search spaces into small diverse pockets of networks. The shrunk search space outperforms the original across a variety of Neural Architecture Search (NAS) benchmarks including achieving 77.6% top-1 accuracy in ImageNet with best in class Kendal-Tau, architectural diversity, and search space size (IJCAI ‘23)

Surveyed the domain of NAS for super resolution to render GPU content at lower resolutions and used ML models to upscale to the target resolutionCreated a novel one shot NAS pipeline for an internal dataset and deployed the pipeline for product applications

Spearheaded a persistent cache for a distributed correlation engine that detects outages in the Google Fiber network which reduced false alarmsDeveloped a server application for Google Fiber that handles outage detection/tracking and many more functions.

Implemented new audio compression algorithm on Google Assistant servers and devices resulting in reduced bandwidth and latency by ~200ms with through testing, experimentation, validation, and characterizationCharacterized and optimized on-device machine learning speech to text models to improve latency across Nest devices

Project Manager for the Dawn team which develops a desktop application built on Electron using NodeJS. Responsibilities include managing a team of 6 developers and performing tasks such as recruiting, handling feature requests, and task distribution and assignment.

Lead of Development Operations team responsible for various projects including: Continuous Integration, Git and Workflow Management, Deployment, Installation scripts, and Beaglebone Black setup.

Created a simulator in python that allowed testing of UI and runtime code without the need for a physical robot. Developed and tested the method for distributing software updates to the many teams that participate in the robotics completion.

The Learning Traffic lab utilizes Deep Reinforcement Learning to apply controls on autonomous vehicles in a mixed fleet of human drivers to improve traffic flow and relieve congestion in simulation. My project involves creating a stochastic model for lane changes and utilizing the model to study the effects of lane changes on traffic and increase the complexity of the model used by the autonomous vehicles.

Implemented machine learning and A.I. algorithms for static code analysis in order to improve efficiency in detecting and preventing bugs. Developed statistical analysis tool in python to process defect records from a bug tracking database queried using SQL. Setup a LAMP application on a Linux server and developed a feature in a Python based internal development application.

Developed firmware for an FDA class III implantable medical device for Spinal Cord and Deep Brain Stimulation in C++ that identifies and configures flash chips with different memory architectures installed on the same PCB. Exceeded project goals of reading and writing to the flash chip by performing auto-detection and configuration of the chip. Participated in firmware code reviews and design reviews.

Lead team meetings, increased engagement through introduction of STEM, website design, creation and administration, establishing Facebook and Google Plus presence.

Created Computer Aided Design (CAD) of the robot, vendor management for manufacture and design, mentor junior team members. Skills acquired/learnt: computer aided drafting and design, programming in C++, and mechanical engineering.