Seattle, Washington, United States
I am the technical (ML) lead for AWS's agentic AI hiring solution (Connect Talent). Learn more here - https://aws.amazon.com/products/connect/talent/ Previously, I led ads in Rufus - Amazon’s agentic AI powered shopping assistant. I have 13+ years of experience in data science and machine learning, including 6.5+ years of industry experience with deep technical expertise in generative AI, agentic architectures, autoML, time series forecasting and semi-supervised learning -- supported by 20+ publications and 300+ citations. Regular contributor and reviewer for top ML and AI conferences -- my Google scholar profile: https://scholar.google.com/citations?user=rX_qvDYAAAAJ&hl=en I enjoy collaborating with diverse cross-functional teams, learning from different perspectives and experiences, and using the latest in AI research to solve pressing business problems. Professional Highlights ➢ I lead ads in Rufus - Amazon’s agentic AI powered shopping assistant. Launched 3 new ad experiences in Amazon’s shopping assistant (Rufus) - product recommendations, brand curated product collections and sponsored prompts (AI-generated contextualized clickable ads), generating $ XX MM USD in annualized revenue for Amazon ads. ➢ Developed 3 LLM-based functionalities over 3 years for Amazon Q (Sagemaker Canvas), leading to 3X% YoY revenue growth and $XXM ARR as of 2025. In particular, the GenAI launches helped increase the end-to-end workflow completion within Sagemaker Canvas, increasing the ratio of completed jobs from 77% to 92%. ➢ Designed, developed, deployed and maintained the evaluation system for the advertising inventory forecasting tool on Amazon advertising’s demand side platform. This was a key component of the bi-weekly business reviews presented to the org-leadership and other product stakeholders. ➢ Worked with NASA Research Center to build models for identifying rare classes when there is a complete absence of expert labels. Applied these techniques on spatio-temporal data from satellites and produced a more reliable and comprehensive burned area database compared to the state-of-art NASA product NOTE: opinions shared on LinkedIn are my own and do not express the opinions and views of my employer.
Senior Scientist for AWS's agentic AI hiring solution (Connect Talent). Learn more here - https://aws.amazon.com/products/connect/talent/
I lead a team of 6+ talented scientists and engineers responsible for building Gen-AI driven conversational ad experiences on Amazon (search, Rufus, product detail pages). • Work on LLM post training, reasoning, personalization and agentic architectures for shopping and ads • Delivered 5+ launches covering Rufus and Amazon Search resulting in $XX MM USD in annualized revenue for Amazon ads.
Developed 3 LLM-based functionalities over 3 years for Amazon Q (Sagemaker Canvas), leading to 3X% YoY revenue growth and $XXM ARR as of 2025. In particular, the GenAI launches helped increase the end-to-end workflow completion within Sagemaker Canvas, increasing the ratio of completed jobs from 77% to 92%. - Designed and developed a generative-AI based assistant data scientist for AWS Sagemaker, launched at re:invent 2024. This chat-based application is able to take your business problem, map it to a machine learning task and walk you through the required data pre-processing and model-building steps recommended for solving that task. The system automatically determines the pre-processing steps required for your data set and the best model for your task. - Designed and developed a chat-based data preparation tool within AWS Sagemaker, launched at re:invent 2023. We supported several use cases including code-generation for data transformations, SQL-generation for Q&A on the data, detecting irrelevant queries, code-generation for data visualization. - Added support for fine tuning open source deep learning models for image and text classification within AWS Sagemaker, launched during re:invent 2023. Built a library over Pytorch Lightning to enable fine tuning with different configurations including Parameter Efficient Fine Tuning (PEFT). https://aws.amazon.com/blogs/aws/use-amazon-q-developer-to-build-ml-models-in-amazon-sagemaker-canvas/
Forecasting advertising inventory on Amazon advertising's Demand Side Platform (DSP). - Designed, developed, deployed and maintained the evaluation system for the advertising inventory forecasting tool on Amazon advertising’s demand side platform. - Presented an overview of the metrics including a deep-dive of problematic forecasting units during biweekly business reviews prepared for leadership and product stakeholders.
Developing machine learning solutions for global-scale land cover mapping problems like creating accurate maps of burned forests. Teaching assistant for Data Mining course taught by Prof. Vipin Kumar