Souvik Ghosh Email & Phone Number

Mountain View, Ca, Us

Sunnyvale, Ca, Us

- I made 800+ ML model compliant with DMA. This involved understanding the law, defining what compliance means for LinkedIn, creating and executing on a plan with ~300 engineers.- [In progress, June 2024] I am leading the large scale AI modeling efforts at LinkedIn. I am currently focussed on building a foundation model to power deeply personalized matching in the LinkedIn marketplaces. This fundamentally involves building a model that understands content and the LinkedIn Economic graph.

Sunnyvale, Ca, Us

I led a 100+ team responsible for researching and developing AI technology for LinkedIn. We build solutions for Personalization, NLP, Multimedia, Federated Learning, Optimization and improving productivity. These technologies are now core to search and recommendations products at LinkedIn like Feed, Ads, Jobs search and recommendations, Notifications and People You May Know and have delivered major business wins including 10% improvement in Ads CTR.

Sunnyvale, Ca, Us

AI Tech lead- initially for member facing products at linkedIn including growth, feed and notifications and eventually for all of AI at LinkedIn. Responsible for strategic AI investments for LinkedIn. I lead AI Academy at LinkedIn where we train employees in AI. The Academy is wildly popular at LinkedIn and we have trained more than 2000 employees from all job functions so far.

Sunnyvale, Ca, Us

Tech lead for LinkedIn feed and content relevance- set technical direction, grew talent and built one of the strongest and high-performing teams at LinkedIn. Improvements delivered by the team resulted in record levels of member engagement and was called out 5 quarters in a row in the quarterly financial report for the company.

Sunnyvale, Ca, Us

Tech lead for 'People You May Know' or PYMK. PYMK is the well known graph link recommendation product and is considered to be one of the most important products at LinkedIn. It is also one of the oldest recommendation products at LinkedIn. PYMK was widely considered to be a solved problem internally at LinkedIn. With a talented team of data scientists and engineers, using careful data analysis and statistical modeling, we delivered 40% metrics improvement for PYMK.* During my first 2 years at LinkedIn (May, 2012 - June 2014) I was an integral part of the advertising relevance team. I worked on click prediction problems, optimizing ad serving mechanisms using budget pacing, pricing optimizations and ad auction bid suggestions. My efforts resulted in an increase in yearly revenue to the tune of $50M.

Sunnyvale, Ca, Us

Built a tool for forecasting event based high-dimensional time series data (for example web-scale impressions, clicks). * The forecasting tool achieved good accuracy rates for LinkedIn data.* The forecasts for ad impressions data was used in developing algorithms for optimizing serving ads at LinkedIn and that achieved 5% improvement in ads revenue.

Sunnyvale, Ca, Us

* Forecasting supply of impressions for advertising contracts - very high dimensional time-series data; improved accuracy by 50% using time-series modeling * Bid-landscape forecasting for ad-exchange data - huge complex data; improved accuracy by 40% by combining techniques from time series analysis and extreme value analysis

New York, Ny, Us

I focussed on research on topics in statistics and probability including time-series, extreme values, large deviations and spatio-temporal processes.I also taught graduate and undergraduate level courses:* Statistical Inference and Time Series Modeling (Masters and Ph.D. students), Fall 2009, 2010* Stochastic Methods in Finance (Masters and Ph.D. students), Spring 2010, 2011* Introduction to Statistics (Undergraduate students), Spring 2009* Sample Survey (Masters and Statistics Major students), Fall 2008

Doctor Of Philosophy (Ph.D.), Probability And Statistics

Masters; Bachelors, Statistics