Jan 2024 – Dec 2025 · 1 yr 11 mos

• Building Glean AI Assistant to bring Work AI to All!

Jan 2020 – Jan 2024 · 4 yrs · Cupertino, California, United States

• Worked on foundational ML models to help launch Apple's very first MR headset: the Vision Pro!

Apr 2017 – Jun 2017 · 2 mos · Stanford, CA

Jan 2017 – Mar 2017 · 2 mos · Stanford, CA

Sep 2016 – Dec 2016 · 3 mos · Stanford, CA

• Helped teach classes, change the curriculum, designed homework assignments and tests.

Jan 2017 – Jan 2020 · 3 yrs · Menlo Park, CA

• Working in the Computer Vision team in Facebook AI specifically the video understanding team. Some of the projects I have worked on:
• Optical Character Recognition:
• Built the end to end system for recognizing text in videos at facebook scale, used by more than a dozen product teams across facebook.
• Zero to one project that includes large scale infra, training/ making frame based detection and recognition models work for video at production scale, building video models to make the OCR process more efficient.
• Gave a talk at @Scale 2019 conference about this work: https://lnkd.in/gPirUGM
• Localized Video Embeddings:
• Built end to end service to provide rich, temporal representations for videos to be used in various classification use cases.

Jun 2016 – Aug 2016 · 2 mos · Mountain View, CA

• Using Deep Learning for YouTube video classification for better search, recommendation, ads targeting and brand safety. Performing applied machine learning research and making models work at the scale of Youtube.

May 2015 – Aug 2015 · 3 mos · Menlo Park, CA

• Implemented a data framework for admin messages in messenger web backend:
• Designed and implemented an easy and efficient API to create, change and access admin messages
• Converted all previous admin messages to implement the framework making changes through the web stack
• Refactored the mark read pipeline:
• Full stack work making many incremental changes to make the mark read and read receipt operations pipeline code in messenger cleaner, more performant and designed to make many planned future changes simple

Aug 2014 – Nov 2014 · 3 mos · San Francisco Bay Area

• Automated aspects of training object models through scripts; trained objects models; used parts-based model algorithm to train object models
• Developed web features for an object recognition project
• Implemented a search feature to search on salient color of image instead of text; learned search engine architecture
• Wrote large scale map-reduce jobs using Cascading to understand consumer data
• Implemented a pipeline re-ranking recommendations through user feedback data such as clicks, repins; project included offline service with multiple data jobs to gather, aggregate and load data and online service to real-time rerank recommendations

May 2014 – Aug 2014 · 3 mos · Raleigh-Durham, North Carolina Area

• Developed features for Microsoft's version control system, TFVC, to enable versioning for shelvesets - a set of pending changes not in the central repository
• Designed the changes with consideration to compatibility and efficiency
• Created RESTful APIs for clients to access server functionality
• Implemented stored procedures and server side logic in TFS
• Implemented command line tools and wrote client side logic for Visual Studio

May 2013 – Nov 2013 · 6 mos · Raleigh-Durham, North Carolina Area

• Worked with various business units to analyze requirements and design;
• Wrote and executed test cases for web-services, Mainframe transmissions (RUMBA) and other internal applications in QC 10, ALM 11;
• Created test plans for black-box testing, end-to-end testing, acceptance testing; helped outline UAT testing; filed defects, worked with developers to debug; performed validation after implementation;
• Used QTP to automate data generation; used SQL to find test data