Oct 2019 – Sep 2024 · 4 yrs 11 mos

• Prompt Optimization
• Developed and implemented OPRO and Protegi algorithms to optimize prompts, reducing token
• usage by 53% and enhancing model performance for defensive search and content harm detection.
• Presented tools to the TREX team, facilitating organizational adoption and improving prompt efficiency
• across Bing.
• RAG Fanout Pipeline for Bing Chatbot
• Built a pipeline using RAG technology to detect jailbreak attacks, scanning over 20 million chat logs
• daily with a 98% precision.
• Enhanced Bing’s protection against jailbreaks, including election-related information leaks, by
• leveraging custom fine-tuned open source mistral Foundational models, reducing costs on large
• language models.
• Fact Check Pipeline
• Created a proactive Fact Check pipeline that generates 300-500 defensive search queries daily with
• 98.4% precision, gaining over 1M impressions since May 2023.
• Stored key queries in Object Store to prevent harmful content, using ChatML prompts and LLMs to
• drive this system.
• MSN Article Reclassification
• Reclassified MSN articles to the travel vertical with a 95% precision, achieving a DAU growth of 350k+
• for travel and 500k+ for health content.
• Collaborated with the MSN team to integrate and deploy efficient classification models for both travel
• and health verticals.
• Underside Travel URL Mining
• Mined 1.4M travel URLs from SLAPI logs with 95% precision, mapped to 40k travel hubs, powering
• underside experiences on Edge browser.
• Mentored an intern to co-develop this work and expand travel-related search coverage.
• Triggering Travel Experiences on MSN Articles
• Designed and deployed a sentiment classifier with 96% precision to rank and tag travel articles based
• on destination relevance.
• Enabled more personalized travel content by associating articles with top-ranked destinations,
• improving reader engagement.

Jan 2018 – Sep 2019 · 1 yr 8 mos

• Built and shipped a neural network-based intent mining solution to improve reminder intent understanding for Cortana Voice Assistant, successfully reducing failure rates from 13% to 4.4%!
• Designed and delivered a Proof of Concept for ConceptGraph, an architecture that iteratively learns relationships between entities and expands on the Knowledge Graph's ontology. This innovation was featured in products like Desktop Tabs and Concept Entity Panes.
• Developed an extensible pipeline leveraging ConceptGraph to generate Desktop Tabs for 500K entities, using insights from Bing query logs to surface top intents. Collaborating closely with the Entity Tabs team, we reached a milestone of 5M daily impressions and 8K DAUs!
• Extended the pipeline to support data extraction, helping identify high-demand intents to enhance the Knowledge Graph’s accuracy and relevance.

Jul 2014 – Dec 2017 · 3 yrs 5 mos · Bengaluru Area, India

• I am passionate about applied machine learning and solving business problems with the state of the art ML algorithms. My skills include NLP, ML classification algorithms , probabilistic modelling and Deep learning concepts & basic applications. Engineering skills include problem solving and data structures & Design.

Jul 2012 – Aug 2013 · 1 yr 1 mo · Hyderabad

• Built a complete website based on content management system Joomla .
• Written real time crawlers for e-commerce websites

Jan 2012 – Apr 2014 · 2 yrs 3 mos · Hyderabad Area, India

• Worked as a Teaching assistant for various courses
• Currently Teaching assistant for C programming and Data structures

Nov 2011 – May 2014 · 2 yrs 6 mos · Hyderabad Area, India

• Working on consistency of the annotated corpora while using annotated corpora for various NLP applications

May 2011 – Jul 2011 · 2 mos · IIIT Hyderabad

• Developed a Web portal in PHP