Led the development and maintenance of high-quality, highly available AI services, managing and coordinating the work of geo-distributed teams across functions (application, data science, AI/ML, and platform engineering), on Salesforce’s next generation AI/ML platform built on AWS.Built the Responsible AI engineering team and led the development of the Trust Layer for Einstein GPT, the world’s first generative AI for CRM. Trust layer capabilities included data masking (detecting and redacting PII & payment data), prompt defense, and recognizing various kinds of harmful & toxic content going to/from target LLMs, powering highly personalized, multi-lingual, retrieval-augmented, generative experiences across Sales, Service, and Marketing applications.Partnered with Salesforce's Office of Ethical and Humane Use, Product Management, and Research teams to define and refine the strategy and roadmap of the Einstein Trust Layer, including providing expertise on buy-vs-build evaluations of vendors focused on the Responsible AI domain.Hired, mentored, and retained top-talent across management, engineering, and science functions in a fast-paced organization, growing dynamic teams capable of delivering quickly and reliably on established and emerging opportunities spurred by Generative AI.Led development of the Salesforce Builder applications (Einstein Prediction Builder, Einstein Recommendation Builder) on top of Salesforce’s next generation AI/ML platform, enabling customers to build predictive and recommender applications on their Salesforce data via no-code tools at a fraction of the cost-to-serve.

Led a geographically distributed team of data scientists & ML engineers, partnering with product and application engineering teams to define the vision & roadmap, and to build Einstein Search, the next gen search system powering Salesforce Search, the most used feature across Salesforce’s platform and products. Indexing hundreds of billions of documents, fielding millions of queries a day and running on thousands of servers worldwide, Salesforce Search is arguably the largest enterprise search implementation in the world.Responsible for building and productionizing various machine learning models for ranking, personalization, natural language search, query suggestions, autocomplete, question-answering, time-series forecasting & anomaly detection, with many of these algorithms integrated into ml4ir (https://github.com/salesforce/ml4ir - a Salesforce sponsored, TensorFlow-based, open-source library for training and deploying deep learning models for search applications).Built the analytics infrastructure that underpins Einstein Search’s user interaction log processing systems enabling the team to understand how our customers use our products and to feed those signals back into relevance systems for modeling and evaluation.Led the team responsible for metrics design, instrumentation specification, design and analysis of A/B experiments, data pipelines & repositories, and visualization dashboards. Drove automation of various aspects of the Search ML infrastructure associated with ideation and productionization of relevance innovations leading to 10x, and in some cases 100x, improvements in rate of delivery of models and features for A/B testing, analysis, and production deployment.

Helped build Salesforce Einstein (http://einstein.com), a layer of artificial intelligence integrated in the Salesforce Platform that delivers predictions and recommendations based on customers’ unique business processes and data. Contributed to https://transmogrif.ai (a Salesforce sponsored open source end-to-end AutoML library for structured data written in Scala that runs on top of Apache Spark).

Led an agile team of data scientists and engineers to incubate data products based on LinkedIn’s massive datasets; initiatives in the areas of social network analysis & visualization (InMaps, OrgMaps – visualizations of members’ professional and organizational networks and applications to inferring organizational structure), reputation (system described in paper: “Personalized Expertise Search at LinkedIn” published at IEEE Big- Data 2015, best paper award), and insights derived from LinkedIn’s Economic Graph (analyzed international migration trends of professionals, paper “Migration of Professionals to the U.S. - Evidence from LinkedIn data” published at SocInfo 2014).

Built learn-to-rank machine learning functionality into LinkedIn’s Lucene-based search and recommender systems, integrating various heterogenous signals at multiple ranking levels under tight latency requirements.Developed node classification and ranking models for large-scale graphs. Applications included modeling the seniority of LinkedIn members and modeling the relevance of LinkedIn profiles to skill keywords. These models were integrated into LinkedIn’s search products to produce lift in CTR and various downstream metrics. Patents granted in the domain of reputation systems.Developed a framework for performing multiple-objective optimization in ranking of recommendations, used to identify Pareto optimal solutions throughout the LinkedIn website to balance various click-through and downstream metric combinations. System described in “Multiple Objective Optimization in Recommender Systems” paper published at RecSys 2012.Modeled the job-seeking propensity of LinkedIn members by performing survival analysis on millions of employee positions, used to power many aspects of the LinkedIn user experience, patent pending.

Designed and built the analytics infrastructure for Suncoast from the ground up (data warehouse, ETL, and reporting) and leveraged this infrastructure to build Suncoast’s first predictive application, built around a model that estimated the probability of default for each existing loan account.This work was done at the height of the 2007 – 2008 financial crisis, when identifying loans that were likely to default became particularly important. Suncoast was, at the time, the largest credit union in Florida and the 8th largest in the United States (ranking by assets).