Sep 2022 – Aug 2024 · 1 yr 11 mos · Menlo Park, California, United States

• ML Engineer on Instagram Feed Ranking Team which generates the feed for the homepage of Instagram, the surface where IG users spend most time.
• Responsible for generating the final feed after blending different types of content together: images, reels(short videos), carousels(more than 1 photo/reel post) and different types of posts together: connected, unconnected (recommendations), hashtags

Mar 2022 – Sep 2022 · 6 mos · Mountain View, California

Dec 2020 – Feb 2022 · 1 yr 2 mos · Mountain View, California

• Feature lead in building the Word Pronunciation Critique for Presenter Coach which listens to user audio and flags incorrect word-pronunciations in real-time.
• This ML tech also powered the “Reading Progress” feature in Teams which has been named one of the “Top Ed Tech Products of The Year 2021!” and has received the “Best Product” award at the Future Education Technology Conference.
• 2 patents filed with USPTO, pending approval.

Sep 2019 – Dec 2020 · 1 yr 3 mos · Mountain View, California

• • Developed a Filler Phrase detection model for Coach which flags use of excessive filler phrases (e.g. you know, actually, basically) while rehearsing. The model is currently deployed in Production for Presenter Coach and Meeting Coach.

Sep 2018 – Sep 2019 · 1 yr · Mountain View, California

• Internationalization of NLP features in Designer to top 20 languages (based on traffic) other than English. Because of the language expansion work, percentage of supported requests (Coverage) of PowerPoint Designer went up from 58%->98% and #UserKepts (no. of times user applied a Designer suggested design) went up by 80%. Patent filed for this work with USPTO, pending approval.
• Built first of its kind Voice Assistant in PowerPoint for linear, non-linear, content-based navigation. Implemented Intent detection from speech transcriptions without wake-phrases such as “Alexa” or “Okay Google”.

Jul 2017 – Sep 2018 · 1 yr 2 mos · Mountain View, California

• Built a Computer Vision service from scratch running variety of Vision models for Face Detection, Salient Object Detection, OCR and Image Classifiers.
• Own the service end-to-end including scalability, performance, telemetry, and deployments.
• Service handles more than 100M requests per day coming from all the images uploaded in PowerPoint daily.

May 2016 – Aug 2016 · 3 mos · San Francisco Bay Area

• Worked on Send Time Optimization problem that enables firms to identify the best time to launch email/push campaigns to target maximum customers, leading to more user engagement, more open rates and click rates and in turn increasing their conversion rate.

Aug 2015 – May 2017 · 1 yr 9 mos · Atlanta, Georgia

• Graduate Teaching Assistant for the course Automata and Complexity (CS4510).

May 2014 – Jul 2014 · 2 mos · Bangalore

• Used concepts of Sequence Mining (support and confidence) and Association Rules to build two tools - an Analysis tool and a Prediction tool, to solve an age old problem in marketing of finding the best customer touch sequence that leads to their conversion for the firm.
• 2 US Patents granted for the novel solutions invented to solve this problem.