Jan 2013 – Jan 2015 · 2 yrs · Ithaca, New York Area

• Robo Brain learns and shares knowledge representations from WWW, leading research groups in order to enable robots to carry out a variety of tasks. DSLL explored different machine learning techniques at scale to build a holistic model by looking at multi-modal social data flowing through different networks and inferring using joint representation.

Jan 2012 – Jan 2016 · 4 yrs · Sunnyvale

• Predict Effect is a marketing intelligence and audience management platform powered by Machine Learning. PE partnered with several large advertising entities in the direct response space and built the first algorithmic cross channel ad optimization engine by embedding all their audience activity into a social collective intelligence space and extracting meaningful insights.

Jan 2010 – Jan 2012 · 2 yrs · San Francisco Bay Area

• We are responsible for building highly scalable, available, fault tolerant infrastructure services and tools to support Netflix's footprint on amazon aws. Select projects involved in:
• Chaos Monkey: Developed the first chaos monkey to ever run on all Netflix production instances. Also extended it to conformity monkey, janitor monkey and laid foundations for simian army framework.
• Actively involved in making Cassandra as our standard data store on the cloud by interacting with several teams across the company to understand their data store needs and simultaneously working with Datastax.
• Co-developed Priam that runs along side of all our 50+ multi regional Cassandra clusters to provide a centralized configuration management, automated token assignment, consistent backup and point in restore functionality, and RESTful monitoring metrics.

Jan 2008 – Jan 2010 · 2 yrs

• Worked on Data Highway, a realtime data platform that collects and analyzes 300 Billion online events (10TB) a day generated across all Yahoo properties with a total hardware installation footprint of 500,000 nodes. The analytics generated are used for Behavioral ad targeting, Personalizing News Feed, Business Intelligence & Key revenue bearing reporting (sponsored search ad clicks, display ad views).
• You Rock award & Data Wizard winner for outstanding performance in the cloud computing division of Yahoo.

Jun 2007 – Sep 2007 · 3 mos

• Worked on identifying the class priors, feature probability distributions for classifying Adult/Non-Adult and Spam/Non-Spam web pages in the vertical search space.
• Also experimented with different SVM kernels, kernel parameters and soft margin parameters.
• Implemented both Bayesian and SVM based classifier with the above identified features which got adapted by several Yahoo verticals.

Sep 2007 – Jun 2008 · 9 mos

• (Advisors: Prof. Hector Garcia Molina, Dr. Andreas Paepcke)
• Web page near duplicate detection: Worked on spotsigs, a new way of computing the signature of a given document that helps to detect duplicates or near duplicates on the web and huge text corpora.
• Crawler snapshot difference: The Stanford webbase project has been collecting topic focused snapshots of websites for a long time. I worked on effective ways to find the difference of two crawler snapshots that helps to build tools on the top of the archives to figure out interesting patterns by historians, sociologists and public policy professionals. It also helps in optimizing features like scoring the activity on a domain, frequency of crawling etc.

Jan 2007 – Jun 2007 · 5 mos

• (Advisor: Prof David M. Golden)
• Worked on Prime Kinetics team which is responsible for developing predictive models of chemical reaction systems that is based on the scientific collaborator paradigms. This project has active involvement from Stanford, MIT and UC, Berkeley. I was responsible for the designing the data model and backend database infrastructure.