Jan 2020 – Mar 2021 · 1 yr 2 mos

Feb 2017 – Dec 2019 · 2 yrs 10 mos

Jul 2014 – Feb 2017 · 2 yrs 7 mos

• Watson Knowledge Graph -- Making sense of approximately 500 million facts known to Watson's Rain Man
• Helping Watson understand what users are looking for:
• query understanding
• entity disambiguation
• fact ranking
• knowledge discovery

Aug 2018 – Present · 7 yrs 7 mos · New Delhi, Delhi, India · On-site

Jul 2013 – Jul 2014 · 1 yr

• Working with a team of researchers developing a novel social media analytics platform. Focus is on developing
• novel tools and services to help business clients utilize social media to reach and engage with target consumer population.

Feb 2013 – Apr 2013 · 2 mos · Webster, New York, United States

• Responsible for developing models for detecting customer complaints by analysis of user tweets. Also developed an algorithm for extracting relevant portions of text from long social media documents like facebook
• posts, blogs, etc.

May 2012 – Aug 2012 · 3 mos · Redmond, Washington, United States

• Designed and developed IM an Expert– a real-time, social question answering service that allows users
• to quickly and easily find answers by asking other users of the service. Using Facebook as the underlying
• social network, I developed a virtual Facebook user that connects the question asker (a Facebook user)
• through Facebook chat to other Facebook users that may potentially know the answer to asker’s question.
• Users’ Facebook profile information and social activity is utilized to create their expertise profile. Also
• designed and developed the experimental framework to study the user interactions.

Sep 2011 – Dec 2011 · 3 mos · Palo Alto, California, United States

• Work involved analysis of query logs to discover various possible intents a given user query may represent
• and use these to produce a diverse set of search results. Developed an algorithm to utilize query
• reformulations and user sessions to produce diverse search results.

May 2011 – Sep 2011 · 4 mos · Almaden Research Centre, San Jose, CA

• Queries entered by users in many practical applications like question answering systems are much longer
• than standard ad hoc information retrieval queries (2–3 keywords). Such verbose queries require special
• processing to uncover the underlying information need of the user. I developed and implemented a scalable
• algorithm for retrieving results for such verbose queries. Achieved 700% faster response times
• while maintaining the quality of search results.

May 2010 – Aug 2010 · 3 mos

• For many search applications like Enterprise search or Intranet search, the user base and hence, the resulting
• query log data are small. As a result, existing query suggestion techniques that require such data can not
• effectively be used in such scenarios. I developed a query log oblivious Probabilistic Query Suggestion
• mechanism that generates query suggestions from the corpus and does not depend on query logs.