Sep 2023 – Present · 2 yrs 6 mos

• Leveraging LLMs to generate query plans from user utterances to help retrieval augmentation through semantic search, knowledge representations, and graph learning, for synthesizing output to the user chats with M365 Copilots. Excited about creating the next generation of human-computer interfaces.

Mar 2019 – Sep 2023 · 4 yrs 6 mos

• Leading teams responsible for various facets of Microsoft Search, Assistance and Intelligence by building training deep learnt models over petabytes of email, calendar and communication data and serving it at scale to hundreds of millions of office 365 users. The team range of features varying from enterprise search, personalized knowledge graph creation, hosting of deep learnt models over CPU and accelerated hardware, Communication Intelligence features like Smart Reply and Task Extraction, Speller and Query Understanding in Search and Email and Calendar management skills for Microsoft Assistant.

Sep 2012 – Mar 2019 · 6 yrs 6 mos

• I have been leading teams responsible for running Bing speller and language/intent understanding services which are utilized by Bing, Cortana, Office/Exchange and even third parties. Each of the services involve building cutting edge machine learning algorithms. web-scale datasources, large scale ML training, query data mining, slot filling, deep learning and statistical translations, as well as building distributed systems at scale for both the run time systems & offline tools for experimentation, metric computation, training and debugging. I have mentored several new technical/people leaders in the process as well.
• Hiring engineers across all levels for our query intelligence team: https://careers.microsoft.com/jobdetails.aspx?jid=267105
• Other job openings in our group: https://careers.microsoft.com/search.aspx?p0=XXOIRecruitXX#&&p2=all&p1=all&p3=all&p4=all&p0=XXOIRecruitXX&p5=all

Jun 2010 – Sep 2012 · 2 yrs 3 mos

• Led a strong team on a mission to re-build the scale of Bing speller to create world's most accurate and comprehensive spelling correction system via a metrics driven engineering culture and scaling across the board, from the lexicon sizes to language/translation to our fuzzy search capabilities. We also made deep investments in breakthrough technologies toward the problem with the help of Microsoft Research.

Oct 2005 – Jun 2010 · 4 yrs 8 mos

• Worked on building a C++ based distributed system that could process, clean and load the entire web graph to run PageRank like graph algorithms for crawl prioritization, index selection, index freshness, spam reduction and developing query dependent ranking features for web search. In this time period, we made several strong innovations in graph algorithms to improve the index quality metrics on one hand and performance/reliability of the run on the other.