Oct 2017 – Sep 2018 · 11 mos

• Egret : An Inmobi initiative platform which enables real time stream joins/transformations. We started it from scratch in a team of 2. After the project matured I am acting as a gatekeeper of the same. Mentored engineers to on-board on the project. Main focus was of providing a distributed environment solution to achieve real time enhancement of data while maintaining low latencies. Zero data duplication and data- loss were achieved in production.
• Azure POCs : Worked on various POCs for tech stack migration on azure cloud.
• Benchmarking BLOB vs ADLS on azure as an alternative for storage.
• Benchmarking performance of HIVE vs SPARK on running heavy queries (on 4 TB data) in azure.
• Designed and created the deployment tool which brings up an on-demand storm cluster in azure for various teams in Inmobi.
• Elastic Storm Service : Working on creating a service which will auto-scale a storm cluster according to incoming traffic. Main challenge is to achieve a sweet spot between horizontal and vertical scalability automatically.

Feb 2016 – Oct 2017 · 1 yr 8 mos

• Egret : Part of team of 2 which started development of platform that enables join/merge/transformation of continuous flowing streams of data (Kafka). This project aimed at making ETLs MR free in local DC. Storm is used as a processing engine.
• Egret is deploying and maintaining all local dc pipelines which ingests around 800 M events per day in a data centre. It is easily scalable and we have faced zero data-loss and data duplication issues in production.
• On making the pipelines MR free, egret has reduced the waiting time of services and business team which are waiting for the enhanced data. Also it has reduced the CPU requirement to run the pipelines by 40%.
• Designed and developed monitoring, deployment logic of various pipelines running under Egret which are crucial for day to day business of Inmobi.

Jul 2012 – Jan 2016 · 3 yrs 6 mos · Bengaluru Area, India

• Aerospike is the only real-time NoSQL database built from the ground up to take advantage of today’s multi-core, multi-processor architectures and solid-state drives (SSDs). That’s what makes it fast – 1M TPS per node with sub-millisecond latency. Hybrid architecture with indexes in-memory and data either in-memory or on flash. Strong consistency, tightly coupled clusters (auto replication, re-balancing, cross data center synchronization) reliably store billions of objects and terabytes of data. Native flash means fewer servers, less power, lower ops costs than with just DRAM. Deployed in demanding environments for almost 3 years with no downtime.

Jan 2012 – Jul 2012 · 6 mos