I work on the Ads Query Understanding team, where I develop state-of-the-art machine learning models and infrastructure for shopping-related ads on Google Search. As part of the query understanding team, I create ML models that capture query intent, which are used by downstream applications, including retrieval and auctions, as a prefilter to allow only commercial intent. My technical expertise includes C++, Python, and TensorFlow.

Key Achievements:

Innovative LLM Integration: Introduced Large Language Models (LLM) to search ads query intent, resulting in multiple successful launches that generated $0.X billion in revenue for Google.

Leadership: Designed end-to-end projects, from collaborating with research teams and prototyping to establishing distillation pipelines for serving LLMs at Google scale with extremely low latency. Led a team of engineers to explore improvements in the entire pipeline, scoping work for follow-up launches, including introducing continuous distilled models, exploring different architectures in teacher and student model, feature explorations, etc.

Recognitions and Awards: Received multiple accolades, including the Research Tech Impact Award and special mentions from the SVP of Ads for one of the most impactful launches of Q2 2023. Presented follow-up tech talks across Google Ads.

I was part of the Google Home/Nest on-device machine learning platform team, responsible for developing end-to-end smart features across all Google Home devices (horizontal) and within individual device features (vertical). We built multimodal ML/AI features leveraging vision, audio, and radar, requiring collaboration with Research, Hardware, Frontend, and backend services.

Key Projects:

Sleep Sensing: Collaborated with research teams to help shape the sleep sensing project. Built end-to-end pipelines for radar preprocessing and data collection, developed heuristics for consolidating ML model predictions into sleep schedules, combined radar and audio outputs for detailed sleep summaries, and analyzed external environments affecting sleep. Also contributed to engineering productivity testing.

Audio Detection: Designed the On-device Personalized Learning Sound Inference for sound discovery in personalized sound detection. Optimized the pipelines to reduce CPU workload by creating coarse audio event models and matching them to personalized events as needed. Published a patent for our work.

Smart Framing: Developed active framing for participants in video calls. Worked on logging metrics for usage and prototyped improvements such as face tracking using a face-person pose detection (FPP) model in video calls.

CSE332: Introduction to Data Visualization

• Implemented keyboard shortcuts for Evernote’s all new web platform for performing content formatting and various operations such as new note, adding tags and expanding/collapsing notes.

• Implemented Quick switcher, an app wide context switcher, to quickly navigate and search within notes notebooks and workspaces, and sort them according to the most recently accessed items.

• Frameworks used: ES6, Javascript, React, Redux, redux-sagas, ducks, chai-js

• Conceptualized, designed, developed and deployed mobile application for Rubique’s loan lending platform

• Implemented major projects such as search and filter module, share and earn module, document upload section for

KYC, chatbot for rubique, meeting module

• Implemented entire redesign of the app along with animations and API integration with new back-end system.

• Introduced coding style and standards, coding review, QA lifecycle and other industrial grade methods to the

application development, to reduce project turnaround time.

• Implemented several custom libraries, such as custom wrapper around volley, custom seekbar, custom collapsible

calendar class, custom animation classes, etc

• Technologies used: Firebase, REST API’s, Volley, Glide, etc.