Joined Pinterest as one of the early members of the Ads Modeling and Auction Team. I have led many high impact machine learning projects and worked in all parts of the Ads funnel including relevance, ranking, auction and retrieval. I have been tech lead for relevance, Uber tech lead of Ads relevance and Engagement modeling. Since Aug 2022 have been reporting to the head of Ads ML leading some key performance initiatives across M10n. We have a SOTA Ads funnel capable of ranking billions of ads under a tight latency bound.

As the first ML hire at a 40 engineer startup, I built the entire ML ecosystem for training and serving from scratch. Led multiple machine learning initiatives at Nextdoor, focusing on growth, monetization, and search. Built ML models for entity extraction, topic assignment, business name resolution from free text, completely automating business recommendation. Revamped Search by migrating to AWS-hosted ElasticSearch with relevance tuning and personalization. Later, I worked on ad personalization and inventory allocation, driving the growth of the newly launched Ads business.

Developed machine learning models and infrastructure for LinkedIn's Premium subscription recommendation system, which was granted a US patent. Additionally, I designed and implemented an Entity Categorizer framework to classify text content in real-time successfully reducing spam in LinkedIn group discussions by 80%

Developed novel method for fast multidimensional subset scan, for detecting disease outbreaks from hospital data. Research was funded by NSF and UPMC.

Worked on Trending Topic generator for Bing's early experiment with Social Search

Developed Internal Document Search service powered through Apache Lucene in the background

Developed PRASTAVA, an open-source recommendation system utilizing collaborative and content-based filtering techniques. Additionally worked on the problem of identifying citing sentences within research papers with aim to improve scientific paper recommendation

Created Content Cop, a plagiarism detection tool designed for research and patent documents. It utilized advanced web crawling techniques and robust probabilistic matching algorithms capable of detecting paraphrasing and synonyms. Content Cop was recognized as a finalist in the PCQuest India’s Best IT Implementation Awards 2008 and was granted patents in both Singapore and India.