Apr 2023 – May 2025 · 2 yrs 1 mo

Feb 2022 – Apr 2023 · 1 yr 2 mos · San Francisco Bay Area

• Staff Software Engineer, Tech Lead for Tweet Search
• Responsible for ranking and relevance for tweet search stack
• Delivered improvements in social engagement metrics on SRP with various ranking improvements such as improved data pipeline, feature normalizations, searcher-author affinity signals.
• Worked on Explore-Exploit for integrating candidate sources to Tweet Search ranking candidates’ pool

Jun 2020 – Feb 2022 · 1 yr 8 mos

• Working on various deep learning efforts for homepage relevance feed
• Transforming feed models to be fully served by a deep learning stack
• Image understanding for feed ranking and candidate selection

Sep 2017 – May 2020 · 2 yrs 8 mos

• * Worked on building end to end content understanding model pipeline for feed updates with deep learning stack. Pipeline serves content relevance signals to various parts of feed relevance models for homepage ranking.

Sep 2016 – Aug 2017 · 11 mos

• Worked on implementing DSSM (a deep learning based query/document representation mechanism) for job search.
• Improved relevance for long tail queries by improving their candidate selection.

Jan 2015 – Aug 2016 · 1 yr 7 mos · Webster, New York

• o User Modeling from Social Media Content
•  Designed and Implemented supervised topic models for tweets and content organization [Java, Mallet]
•  Implemented methods for entity disambiguation and linking to Semantic Knowledge Base (e.g. Freebase) for users’ interest detection [Java, Apache Stanbol, MongoDB]
•  Designed and implemented multi-modal deep learning based methods for images, text and graph data in users’ social networks for multi-label classification [python, Caffe, Theano, Keras]
•  Designed and implemented natural language based feature extraction methods for Life Event Detection in social media [Python, Scikit-learn, Stanford NLP]
•  Led the development of prototype for Real-time Interest Profiling engine combining above modules.
• o Leveraging Natural Language Processing for Patient Profiling and Predictive Health Modeling
•  Implemented deep learning based Neural Reasoning models to extract (embedding) features out of textual and semantic knowledge sources (e.g., PubMed/MedLine, UMLS, etc.) [python, Theano, cTakes]
•  Worked on methods for patient profiling and disease progression from medical forums (such as healingwell, WebMD) for various chronic diseases. [python, cTakes]

May 2012 – Dec 2014 · 2 yrs 7 mos · Webster NY

• o Data Mining for Customer Care Center Optimization
•  Analyzing Agent and Customer Conversations: Researched and developed multi-task learning techniques for analyzing conversations/logs. Implemented various machine learning techniques for optimizing customer satisfaction such as dialogue act classification, detecting issue resolution status, call center agents’ efficiency. [Mallet, Scikit-learn, Java, Python]
•  Linking Users across CRM databases and Social Media: Designed and implemented machine learning based techniques for linking users across their corresponding CRM database transactions and social media accounts as well as linking users across various social media accounts (e.g., twitter, Flickr, and Instagram). Implemented various NLP (e.g. Entity Recognition, personality identifier, Interest detection) and ML (e.g. Kernel Density Estimation, Logistic Regression) based techniques for scoring similar users. [Python, Scikit-learn]
•  Community Detection for Peer to Peer Support: Designed and implemented topic modeling based community detection techniques in social media. The statistical method proposed can distinguish between users’ conversations towards an implicit community in the social network and personal statuses. Co-led the development of user-community recommendation engine based upon the community detection algorithm. [Mallet, Java]

May 2010 – Aug 2010 · 3 mos · Bengaluru Area, India

• Hierarchical Topic Models for Entity Disambiguation:
• Researched and implemented weakly semi-supervised topic models for detecting, disambiguating, and linking entity mentions in textual documents to their Wikipedia profiles. The developed methods outperformed state of art entity linking methods and overcame issues such as context window selection, collective linking in documents, etc. [Mallet, Java]

Jun 2009 – Sep 2009 · 3 mos · Grenoble Area, France

• o Font Image Retrieval: Researched and implemented a retrieval system for font description files that are visually similar to user’s “hand-drawn” font. Implemented various combinations of ML based classifiers and feature extraction techniques. [Matlab, C++]

Aug 2007 – Mar 2012 · 4 yrs 7 mos · State College, Pennsylvania Area

• o Citation Recommendation Engine (RefSeer): Researched and implemented a content based document recommendation system that ranks scientific literature based upon likelihood of being cited with given user’s scientific text. The developed system interfaces with CiteSeerX and extends topic models for linked corpora as its underlying ranking algorithm. [Mallet, Java, Spring Framework]
• o Information Extraction from Scientific Literature: Researched and implemented information extraction modules that extract structured output from 2-D plots of certain classification (e.g., bar chart and line chart) and textual description of figures. The modules are integrated with CiteSeerX digital Library system. [Matlab, C++]

Jan 2005 – Jan 2006 · 1 yr · Noida Area, India