Apr 2017 – Jun 2021 · 4 yrs 2 mos

• Led below initiatives, resulting in significant product impact across Feed, Learning, and Career verticals.
• Standardization of Linkedin member activities through state of the art Transformer based models, and Extreme Classification methods; Platform for various downstream use-cases of unified activity representations across search, recommendation, and notifications in Career vertical.
• Microsoft Turing Universal Language Models for few/zero-shot classification tasks towards i18n content moderation and trust.
• Content modeling for moderation / distribution through Deep hierarchical sequence models, Personalized content embeddings for feed-ranking.
• Job, People, Linkedin Learning Search: Personalized query auto-completion / rewriting, document understanding, retrieval / ranking models. Deep & wide architectures optimized through Tensorflow-ranking.
• Mentoring, evangelize, educate AI / Machine-learning expertise for all of Bangalore Engineering through AI-academy

Dec 2016 – Mar 2017 · 3 mos

• Scaling/extending distributed training of deep neural networks through data/model parallelisms in Tensorflow framework.

Sep 2014 – Dec 2016 · 2 yrs 3 mos · Sunnyvale, California, United States

• Gemini Search:
• Lead multiple projects on recommending queries/ads, and learning-to-rank models for relevance contributing xx% of daily revenue
• Developed large-scale online methods (Factorization machines / MLR) for matching/relevance-ranking of Gemini Search Ads, tail-query optimization, vertical-specific matching strategies.
• Developed and lead the project in search-campaign optimization (keyword recommendation) with significant RPS gains
• Delivered large-scale graph-based query rewriting techniques for matching of BROAD keywords in Sponsored Search with +RPS, +CTR.
• Gemini Product Listing Ads:
• Lead the science/engineering/editorial team on Matching/Relevance, Query/Doc Processing (Intent, Categorization, Semantic-tagging), Ad-ranking for CTR optimization.
• As an IC developed machine learned models (MLR/L2R) for Product offers retrieval, and ranking, with significant rpm lifts in production.

Aug 2008 – Jul 2014 · 5 yrs 11 mos · Bengaluru, Karnataka, India

• Performance Display Ads Exchange - RMX, RightMedia Exchange:
• Responsible for various ranking initiatives through scalable machine learning methods.
• Responsible for model fitting on Hadoop Map-reduce, Feature Engineering, A/B testing, Data/Metrics Analytics for future iterations
• Developed multiple scalable (M/R) hierarchical models for CTR/CVR prediction on large-scale high-dimensional categorical datasets (tall/skinny).
• Delivered significant revenue gains of the order +XX Million USD in production.
• Extended the log-linear models with hierarchical/ temporal smoothing, and matrix factorization on the ad X pub interactions, published at KDD.
• Behavior / Lookalike Targeting:
• Lead the team of cross-org/geo to adopt Flickr daily-habits/interest profile for ad-targeting / feed personalization.
• Short/Noisy text modeling for entity extraction, disambiguation, and categorization
• Roles over years:
• Principal Research Engineer: April 2012 - July 2014
• Senior Research Engineer: April 2011 - March 2012
• Research Engineer: Aug 2008 - March 2011

Aug 2006 – Jun 2008 · 1 yr 10 mos

Jan 2005 – Jan 2006 · 1 yr