I currently work on signals and deep learning models for homefeed ranking.

➢ Conceptualized and led the development of a system to aggregate data and quantify key performance aspects of business logic like bidding&modeling performance, converter/retargeter funnels. Project involved maintaining a sliding window of 1 trillion events from a 1million QPS event stream, efficient compression/storage, inverted indexing using hbase for highly efficient joins, generating daily statistical insights from peta scale data.➢ Tech Lead for TrafficScanner(https://trafficscanner.rocketfuel.com) - Conceptualized, prototyped and led the development of bot-score prediction models to assist with real time bidding process. Project involved behavior profiling, offline modeling, online scoring against 1mm QPS request stream .The model could predict bot-score given user-id and request context and was used for different purposes.➢ Developed a web-scale crawler to crawl urls where rocketfuel could potentially serve ads to fetch critical information about these web pages.➢ Built and mentored Rocketfuel's inventory quality team and infrastructure around it from the ground up. The team was responsible for modeling, forecasting on bid requests to block bidding, shift bidding landscape or evaluate requests differently to effectively handle suspicious traffic, advertising fraud and foresee problems due to legitimate tech issues(eg:- click fraud, block suspicious ip, foresee misrepresented bidrequests etc).➢ Strong contributions on infrastructure side like probabilistic map join for Hive, machine learning UDFs for Hive, client side cache for hbase, javascript code to fingerprint browsers/users.➢ Developed models to block inventory, shift bid-landscape to control impression drop-off (saves the company millions of dollars every year and helped stabilize the margin which was critical before the IPO). Impression dropoff happens when you win an auction(you need to pay $$) and do not get an opportunity to serve the ad.

Was TA for both undergrad and grad algorithms courses.

➢ Worked on impression dropoff analysis(data mining/ML) between rocket fuel bidders and exchanges for potential margin improvements. The work involved machine learning research on different independent variables varying from garbage collection/queuing pauses in adserving, load balance across clusters to publisher domain specific problems with relatively limited data points.➢ Developed machine learning infrastructure to minimize starvation in ad-auctions by automatically regulating bids on some of the poor performing ad/site/exchange combinations.➢ Real time bidding infrastructure to support custom adserving JavaScript for different A/B testing and performance profiling purposes.

➢ Handled product management and development of complete webstack in LAMP.➢ Worked on ETL and university recommendations feature based on link analysis on bipartite graphs.

➢ Worked predominantly on application layer(Network manager daemon) for L2 switches in design, development and code refactoring projects. ➢ Strong experience with memory management/profiling, concurrency in C/C++, software architecture and development.

➢ Worked on tracking multiple objects efficiently in real time using laser sensors.