Nov 2020 – Oct 2021 · 11 mos

• I worked on Kubernetes, GKE and Anthos. I was one of the technical leads for https://docs.cloud.google.com/kubernetes-engine/fleet-management/docs.
• I led the design, implementation, productionization and evangelization of a multi-tenant multi-cluster Kubernetes controller platform at Google. This platform is now in use by 10+ teams inside Google. My team was responsible for the continued operations and development of this platform.
• Previously, I was part of the three person team that added GPU support in GKE and Kubernetes.

Nov 2018 – Oct 2020 · 1 yr 11 mos

May 2017 – Oct 2018 · 1 yr 5 mos

Mar 2016 – Apr 2017 · 1 yr 1 mo · San Francisco Bay Area

• I was part of Cloudflare's Data Infrastructure team. The team was responsible for the end-to-end log-processing pipeline at Cloudflare. I was primarily involved in improving the resiliency of the pipeline. My work helped scale the pipeline from handling 3.5M QPS at peak to handling more than 6M QPS at peak. I started various efforts to improve Cloudflare's internal analytics tools. I also rewrote a major component of Cloudflare's customer-facing HTTP analytics.

Nov 2013 – Feb 2016 · 2 yrs 3 mos · Mountain View, California, USA

• I wore many hats during my time at Qubole. Here's a reverse chronological list of some of my major projects:
• Built a multi-tenant job history service for Hadoop and Spark clusters. This provides the convenience of always-on permanent clusters at the cost of on-demand ephemeral clusters which is one of the major value propositions of Qubole. https://qubole-eng.quora.com/Multi-tenant-Job-History-Server-for-Ephemeral-Hadoop-and-Spark-Clusters
• Worked in a team of two to build Qubole's Hadoop-2 stack. This involved porting more than two years of improvements and fixes made by Qubole from Hadoop-1 to Hadoop-2 and making it production ready. All the new query engines supported by Qubole including Spark and Tez run on this Hadoop-2 stack.
• Added support for running Qubole clusters inside AWS VPC.
• Improved Qubole's monitoring and reporting abilities by adding Ganglia monitoring to the service and implementing a reporting API.
• Built a service to canonicalize Hive queries to identify top Hive workloads running on Qubole. https://qubole-eng.quora.com/Canonicalizing-hive-queries-to-find-top-workloads
• Built a fast-syntax-checker by extracting the syntax analyzer portion of Apache Hive and exposing it as a Thrift service. This reduced the response time for queries with incorrect syntax from >10s to <1s, while adding only a few milliseconds to the response time of queries with correct syntax.
• I also maintained Qubole's Python and Java SDKs.

May 2012 – Jul 2012 · 2 mos · Marina del Rey, California, USA

• Worked as a Research Intern under Dr. Ewa Deelman, USC Information Sciences Institute.
• The project comprised of a study evaluating a peer-to-peer approach towards data sharing for workflow applications running on the cloud, and the integration of the approach into an existing workflow management system.
• The work resulted in a workshop publication.

May 2011 – Jul 2011 · 2 mos · Birmingham, United Kingdom

• Worked as a Research Intern under Dr. Radu Calinescu, Aston University.
• The project comprised of a study evaluating the compute capabilities of virtual machines running on the cloud, and the development of a cloud-based version of a part of an existing software tool for the analysis of Markovian models.