Jan 2023 – Sep 2024 · 1 yr 8 mos · San Francisco, California, United States

Nov 2021 – Nov 2022 · 1 yr · San Francisco Bay Area

• Started and Co-founded Adept AI where the vision was to build useful and intelligent models that can take actions with tools in your browser on behalf of users.

Apr 2021 – Nov 2021 · 7 mos

• Core contributor of the “Attention Is All You Need” paper (https://arxiv.org/abs/1706.03762) that invented the Transformer model. I also led scaling transformer models and extended it to other AI problems like Image Generation(https://arxiv.org/abs/1802.05751) and Computer Vision(https://arxiv.org/abs/1906.05909). Transformer based models can write news articles, poetry, emails, translate one language into another, and even answer questions. It has also been used in code generation, protein folding, speech recognition and many other AI models like ChatGPT, GPT3, AlphaStar, AlphaFold, Co-Pilot, Dall-e etc.
• Launched some of the first successful deep learning based AI models for Question Answering and Text similarity for Google Search.

Feb 2020 – Apr 2021 · 1 yr 2 mos

Nov 2018 – Feb 2020 · 1 yr 3 mos

Oct 2017 – Nov 2018 · 1 yr 1 mo

• Deep Learning research for language understanding and vision (images, videos)
• Research in Deep Learning for Machine Translation, Images and other generation methods.

Oct 2016 – Oct 2017 · 1 yr

• Research in Deep Learning for Machine Translation, Images and other generation methods.

Jul 2015 – Oct 2016 · 1 yr 3 mos

• Research in Deep Learning for Machine Translation, Images and other generation methods.
• Previously worked in Natural language understanding for sentence similarity and question answering systems using end to end deep learning methods.

Jun 2014 – Aug 2014 · 2 mos · Cupertino, California

• ▪ Part of the Global Action Trace - Distributed Computing team under the E-commerce Platform Services.
• ▪ Tracking dependency changes for any service in Amazon over time to report changes in call patterns like new or removed dependencies. Tracking volume of calls between services, to report causes of discrepancy or latency issues in the call chains.
• ▪ Processing was done in a distributed environment to make it scalable to manage millions of service calls made in an hour.

Jan 2014 – Dec 2015 · 1 yr 11 mos · Greater Los Angeles Area

• https://cssl.usc.edu
• ▪ Social media analysis to determine how moral content in social media text can predict real-world behavioral phenomena like group dynamics, decision making and culture
• ▪ Analyzed twitter data during government shutdown (Nov’13) to predict the social distance between users, their rate of activity and the influence each user has on the twitter network. Visualized network structure to show difference in moral values.
• ▪ Developed a NLP suite to implement algorithms like z-label LDA, clustering, classifiers and social media data extractor.

Aug 2013 – Jan 2014 · 5 mos · Graduate Schhol

• ▪ Performance tuning of database using query optimization techniques. Processing student data to generate survey reports and cleaning up redundant data using normalization.

Jul 2012 – Jul 2013 · 1 yr · Pune Area, India

• ▪ Part of the AdServer team responsible for Online AdServing for publishers in real time across display, mobile and video. Included the end to end ad serving process, requesting bids, sending the optimal ad and gathering stats all in realtime.
• ▪ Worked on designing and implementing video ad serving for publishers supporting video inventory.
• ▪ Used Hadoop (Map Reduce) to aggregate pixel and audience data stats and writing it to MySQL. Also, worked on predicting the mobile ad price from Ad networks using different parameters like device, os, time, location, etc.