Build a fast, reliable and standard mechanism to load and serve Keyval Data across Google Ads

• Leading the effort to develop mock scripts to simulate ransomware attacks on a VM. Developed data generators, capable of simulating a VM’s churn such that entropy of the data and the hierarchy of the file-system tree is highly configurable. The attacks are simulated on this data to train ML models in our ransomware detection pipeline at scale. [Patent Pending]• Split tasks into IO/CPU/network bound and pipelined them to improve file-system indexing performance by 35%.• Designed heuristics to figure out files affected by a ransomware attack and gave users the option to recover them.

• Founded and led a cross-functional team to add SLA Management capabilities to Rubrik’s SaaS Platform: Polaris GPS. Architected a generic framework to keep SLA definitions in sync across Polaris running on Cloud and Rubrik clusters deployed across customer data centers around the world. [Patent Pending]. • Designed and developed an automated setup of a secure peer-to-peer direct replication channel between a pair of clusters where a trusted third party acts as the initial authorizer.• Augmented Rubrik’s proprietary file-system and data management service to expose snapshots 2x faster.

Responsible for finding new web pages where significant revenue can be generated by showing contextual ads by trying out Pearson Correlation, linear models, gradient boosting trees, and Naive Bayes text classifier.

Smart Volume Distribution in SOMS• Added smart package volume distribution feature based on relative capacities of each chute. Implemented and tested a fast randomized algorithm with 90%+ accuracy and a slower deterministic algorithm.

Curation of problems and writing editorials for algorithmic programming contests held on CodeChef,HackerEarth and several internal contests in college.