Improving critical ads ranking models with new signals, new model ensembles, new debiasing and calibration

techniques, new pipelines, etc. Also creating new orchestration primitives for AI agents.

• Since LinkedIn's pivot towards Generative AI, I've been leading a variety of projects involving the creation, adaptation, and use of LLMs. For example, I conceptualized and led the development of a generic natural language query understanding capability for search verticals. Before that, I led the team responsible for embedding-based user ranking for "collaborative articles", which are LinkedIn's fastest growing traffic driver, according to the July 2023 MSFT earnings call. Listen to https://www.youtube.com/watch?v=FLP7m0Qltv0&t=1315s at 13:15.

• Led numerous machine learning projects to improve LinkedIn's firmographics.

E.g., see https://engineering.linkedin.com/blog/2021/linkedin-sales-insights--quality-data-foundations-for-smarter-sa

• Led company-wide migration to new industry ontology, affecting hundreds of client teams and over a thousand code repositories.

Architecture, design, implementation, and operation of the contextual bandit machine learning system behind the Promoted Listings business. (Scala, Python, Java, Spark)

Worked on the architecture, design, implementation, and dissemination of the

Decision Service, the world's first cloud service for machine learning from partial feedback.

See https://portal.ds.microsoft.com and https://github.com/Microsoft/mwt-ds .

Data science for applications of speech recognition. (C++, Python)