Tech Lead, Shopping Ads Relevance LLMs

Build SOTA models to predict the relevance of Shopping Ads shown on Google.com

Google Tech Impact Award for developing Gemini based LLMs for Relevance.

• Performed numerical modeling, computational simulation and correction algorithm development for optimizing virtual photomasks using computational lithography.

• Used predictive modeling, statistics, machine learning, and other data analysis techniques to collect, analyze, and visualize large datasets using a proprietary data pipeline and HPC infrastructure.

• Awarded a Distinguished Invention Award by Intel for developing a new OPC algorithm.

• Filed a patent for a ML based technique to optimize via placement in integrated circuits.

Ph.D. Thesis Research: Multiplexed particle manipulation using microfluidics

* Apply nonlinear model predictive control techniques for hydrodynamic trapping of micro and nanoparticles. Implement efficient particle tracking and control algorithms in C++, Labview, and Matlab.

* Research highlighted on Nature Physics and University blog.

* 3rd Prize, Langmuir Presentation Award Session, ACS Colloid & Surface Science Symposium, Harvard University 2016

* Best Speaker Award, CPLC/Biophysics Graduate Student/Postdoc Symposium, UIUC 2016

* Best Presentation, Particle and Droplet Manipulation Session, AIChE Annual Meeting 2015

* $20000 FMC Educational Fund Fellowship, UIUC 2014

• Implemented machine learning and computer vision algorithms using Python, OpenCV and Scikit-Learn to build a framework for detecting tool failure in Caterpillar’s earth digging equipment using dashboard videos

• Performed data annotation and cleaning, outlier detection and trajectory-based filtering on noisy data for improving the performance of the detector, and filed a patent application based on this project

Summer Research Fellowship at UC Berkeley, working with Prof. Rich Mathies's research group. Fabricated and optimized a multilayer glass microfluidic 4x4 microvalve array using photolithography, and successfully demonstrated on-chip labeling for rapid compositional analysis of bioassays with low biomolecular sample concentrations