Tech Lead for ML models powering critical business use cases in Rides & Grocery Retail.* Developed an RL based model for recommending similar items that led to improvement of item replacement rates for Uber from x% to y%+. This unlocked large business value and improved platform reliability.* Led Item Availability modeling that predicts items being out of stock for fulfillment.* Complementary Item recommendations for items that can be bundled in purchases.* Product classification models - core to categorize merchant products into an Uber canonical taxonomy. These models have org wide consumers and key in detecting restricted items (alcohol, tobacco etc.).* Place prediction model that predicts important places from a user's ride history.* Dish understanding models that predict the type of dish, cuisine, proteins and endorsements. * GenAI: LLMs for classification (label space RAG), label generation and fine tuning for downstream use cases.Leading Content Intelligence stream to understand users deeply and surface a personalized storefront.

• Member of Magma team working on core structured search ranking. • Trained boosting and filtering models for 'recipes' search results on Google which is about x% of search traffic.

TechLead for a group of ML scientists to ship ML models in Microsoft Office Products.• Creator of PowerPoint Designer’s Text2Image ML system that recommends millions of images daily to PowerPoint users. NLP Model to suggest relevant images given the text typed on a slide; from bootstrapping to training on user data and exploration strategies.• Built detector models that decide when to surface image and icon content suggestions in PPT designer.• Ranking models for PPT theme recommendation engine that pair an image, font, template and colorpalette together based on user clicks, various context specific, deep learning and personalization features.• Semantic segmentation models to find salient regions in PPT images.• Speaker at Microsoft MLADS Spring 2020 (Machine Learning AI and Data Science) conference with atalk titled - ‘Building complex user experiences by using subjective-objective data analysis’.• Unsupervised learning, document clustering and file ranking for Word and PPT.

Developed for the data warehouse team.1. OpsDashboard project - worked on a job tracking tool which reduced the operational load onData warehouse support engineers by an estimated 2 hours per day. Full ownership for the development of ticketing module from from front-end to back end. Maintained a graph database for the project - reduced the database down time from 4 hours to zero hours for the production service. 2. Redshift Table Migration - helped automate migrating oracle tables to Redshift by ascheduler copying the oracle partitions and throttling cluster load, saving many man hours.

Zonk provides an innovative wireless entertainment solution for the public transport in order to give the passengers a truly unique travel experience making their strenuous journey fun and stress free. Zonk technologies aims to transform boring bus travel and make it a happy fun-filled experience. Our customers will have free access to various infotainment services via their Wi-Fi enabled personal devices including laptops, tablets or smartphones. We developed an application which is very flexible in terms of functionality and operation encompassing almost every media type - audio, video, text and games. The modular setup of our system also includes the integration of multiple additional features such as travel guides, advertisements, user to user chat functionalities or multiplayer games – all accessible through the browser or a single app on the users devices. There are virtually no limits to the range of services that can be integrated on the platform. Our team has researched a lot at tailoring and delivering the perfect package to provide the passengers with a first-class entertainment experience. The media content will cover almost everything from blockbuster movies, TV shows, audio and games to educational documentaries, safety films and electronic magazines.