Mar 2018 – Present · 8 yrs 1 mo · Palo Alto, California

• Our mission is to accelerate AGI advancement and its deployment in the world.
• The bottleneck for AGI progress used to be compute and data. Now, it’s human intelligence.
• We use humans to make AI smarter:
• Turing trains AGI by working with all the major AI foundation model companies to improve model performance and fine-tuning for coding, data analysis, advanced reasoning, problem solving, multi-modality, function calling, agentic workflows, STEM and advanced knowledge work requiring industry domain expertise.
• We use AI to make humans smarter:
• Turing deploys AGI by working with Fortune 500 enterprises and high-growth startups to build custom models fine tuned on proprietary data, co-pilots, agents and custom AI applications that create strategic business value and amplify human productivity.

Mar 2018 – Present · 8 yrs 1 mo

• StartX runs the world's top startup accelerator and founder community for Stanford‑affiliated entrepreneurs. StartX is a 501(c)(3) Stanford-affiliated nonprofit in Silicon Valley that runs one of the world’s top startup accelerator programs. Its mission is to advance the development of the best entrepreneurs through experiential education and peer learning. Since launching in 2010, StartX has supported more than 400 companies and 900 entrepreneurs, from early to pre-IPO stage, working across a wide spectrum of industries.
• To date, StartX has supported over 700 companies including Snap, Lime, Branch Metrics, Life 360, Periscope, Kodiak (NYSE:KOD), Eargo (Nasdaq: EAR). StartX companies have raised over $2.2B with a $5.1M+ average per company funding rate.

Nov 2017 – Mar 2018 · 4 mos · Menlo Park

Jan 2017 – Jan 2017 · 0 mo · Sunnyvale, California

• Joined Revcontent, following the acquisition of my company, Rover where I served as CEO.
• Rover which has offices in Silicon Valley and Mumbai is now a wholly owned subsidiary of Revcontent. More details on the acquisition here:
• https://techcrunch.com/2017/02/23/revcontent-acquires-rover/
• We use state of the art Machine Learning and Natural Language Processing to power the Recommendation Engine that serves over 250 billion content recommendations/month across the top publishers in the US.
• To,
• Help users discover content they love through Deep Personalization (through rich Document and User modeling)
• Help publishers drive increased eCPM/revenues (Machine Learned CTR Prediction)
• Help publishers and content marketers discover their most valuable users (Audience insights, Interest targeting)
• Help publishers drive increased traffic, session lengths, engagement on their domains (Internal Discovery)
• Revcontent is the world’s fastest growing content recommendation network, powering 250 billion content recommendations per month. Revcontent partners with the largest media brands in the world such as Forbes, Newsweek, The Atlantic, International Business Times, and more. Revcontent’s headquarters lies in Sarasota, Florida with global offices including London, San Diego, San Francisco, New York, and more coming soon. Revcontent reaches 92% of US households, according to Quantcast and has been featured in Forbes, The Huffington Post, Fox News, and more.

Jan 2008 – Jan 2016 · 8 yrs · Sunnyvale, CA

• Rover uses Machine Learning for Deep Personalization to help people discover content they love.
• Formerly known as Flipora/Infoaxe.
• Rover has a Consumer business and an Enterprise business.
• Rover Consumer:
• Apps for iPhone, Android and Web with over 40 Million registered users.
• Featured by Apple under Best Apps for iPhone
• Featured on TechCrunch, Recode, VentureBeat, Forbes, Huffington Post etc.
• Rover Enterprise:
• Offers Personalized Content Recommendations to users via widgets on Publisher sites
• Helps Publishers with monetization and traffic acquisition
• Via Revcontent, reaches 97% of US households (per Quantcast)
• Technology:
• Large Scale Modeling of a user's interests through content they engage with
• Innovations in dimensionality reduction of the machine learned models that translate to massive
• speedup with training time, prediction time and compactness of the models (1000X model size reduction, 50X speedup at prediction time)
• Capability to do on-device personalization in a privacy friendly manner
• Identifying users with similar interests and helping them connect
• Rover had raised about $6.5 Million in Venture Capital at the time of the acquisition.
• Investors & Advisors:
• Stephen Oskoui - Venture Partner at Founders Fund
• Gokul Rajaram - Creator of Google AdSense, Facebook, Square
• Hector Garcia-Molina - Professor, Stanford University in CS and EE, Board of Directors at Oracle
• Barney Pell - Founder & CEO of Powerset (acquired by Microsoft)
• Munjal Shah - Founder & CEO of Like.com (acquired by Google)
• Anil Dharni - Founder of Funzio (acquired by GREE).
• Saeed Amidi - Founder & CEO of PlugandPlay TechCenter
• Adrian Weller - Machine Learning at Cambridge University
• Mayank Bawa - Founder & CEO of Aster Data (acquired by Teradata)
• Ilya Fushman - Head of Product at DropBox, Partner at Index Ventures
• Tim Draper - Founder & General Partner at DFJ
• Shariq Rizvi - Director of Ads, Twitter
• Labrador Ventures
• Band of Angels
• Amidzad Partners & others

Jul 2007 – Jan 2008 · 6 mos

• Powerset was a Natural Language Search Engine where the goal was to read and understand every sentence on the Web to answer queries posed by users in natural language. This required deep syntactic parsing of documents & queries, followed by semantic analysis and markup leveraging resources like WordNet, FreeBase etc.
• At Powerset, I co-designed the Ranking Algorithms for the Natural Language Search engine that outperformed Google, Yahoo! and Live Search on Wikipedia on the Discounted Cumulative Gain metric. My work involved coming up with and combining various ranking features that were the result of deep syntactic and semantic analysis of the content of the documents, as well as keyword based and Web graph derived features. I was also responsible for optimizing and improving the ranking function in a systematic way by running several relevance tests on Mechanical Turk and using the results to inform future improvements to the core ranking algorithm of the search engine. I also trained a machine learned ranking function for the natural language search engine.
• Powerset was acquired by Microsoft for a reported $100M and became a part of Bing.

Jun 2006 – Sep 2006 · 3 mos · Santa Clara

• At Yahoo!'s Machine Learned Search Ranking group I worked on Automatic Search Regionalization where my changes resulted in one of Yahoo!'s biggest search relevance gains that year for a single feature. This is the problem of deciding what subset of the query stream needs regional/local content boosting eg. "buying digital camera" or "driver's license", so that the ranking function scores regional/local content more heavily. On the flip side, a query like "neural networks" needs no regional/local content boosting since we want to surface the most globally relevant results.
• I built a Query Classifier for regional intent that used cues from the user query stream and clickstream data and signals from. I then retrained the search ranking algorithm using signals derived from this which gave a significant boost to search relevance (on the Discounted Cumulative Gain metric).