Designed and implemented a C# ad-serving engine for precise ad personalization and extensive experimentation, developing a distributed, scalable, fault-tolerant system for high throughput with automated retry mechanisms. Maintained critical systems, impacting organizational metrics, and led integration efforts across teams with endpoint development and documentation to streamline operations.

Research on domain generalization techniques to improve the performance of models on out-of-distribution (OOD) datasets such as WILDS.Investigation of sampling algorithms for effective training on datasets with non-trivial OOD domains.Development and implementation of group fairness models for MIT-IBM AI fairness user survey.

Led the development of Cloud Init test automation infrastructure to eliminate bottleneck manual tests for crucial cloud OS initialization.Integrated modular testing infrastructure into existing testing code base with no interference.

Researched NVMe tunables and developed adaptable infrastructure to supportautomatic collection of NVMe throughput data with varying parametersDeveloped Deep Q Network based and Genetic Algorithm machine learningmodel to select optimal parameters based on influence on throughput

The UROP works with the industry partner Wayfair, one of the world’s largest online seller of home goods. Wayfair runs many simultaneous promotional events on its website. Each promotional event contains a set of products that are discounted from its nominal price. The UROP seeks to optimize Wayfair’s promotion planning process by jointly making optimal pricing and product selection decisions across different events. In order to answer these questions, our goal is to understand and predict customer purchase behavior, using historical transactional data and state-of-art machine learning tools. The overall goal of the project would be to apply the models and algorithms constructed through the research to see the impact in practice. Machine learning techniques will be utilized to predict customer behavior related to the pricing and promotional items.

Experimental research and data analysis in the ABRACADABRA project which searches for axion dark matter using a toroidal magnet to source an effective electric current to enduce an oscillating magnetic flux to be detected and amplified through a pickup loop and SQUID magnetometer. Development of complex circuit models using SPICE of the ABRACADABRA device to minimize self-interference in order to maximize the precision and accuracy of extremely low-intensity magnetic field measurements.

Work on the development of Python-based Blender toolkits that would assist in the near real-time generation, recording and repetition of complex 3-dimensional paths for the UR5 industrial robotic arm with motion tracking devices. The toolkit will allow for the robot to draw and paint in a high degree of motion control for a broad range of stroke and mark making.