Jun 2022 – Jun 2024 · 2 yrs

• Optimized Kubernetes clusters through Karpenter's advanced bin packing and node provisioning, realizing substantial annual savings of $160,000. Surpassed the goal of a 60% cost reduction across all clusters. Tech Stack Used: Karpenter, Kubecost, Kubernetes, EC2, AWS Fargrate.
• Designed an event-driven infrastructure for ML and AI production workloads for Kubernetes, seamlessly integrating DAPR and Kafka as universal event brokers. This architecture serves as the foundation for successfully deploying over 15+ models into production. Tech Stack Used: Kafka, DAPR, Flask, Gunicorn, Kubernetes
• Implemented dynamic pod scalability using KEDA and a custom gRPC server for workload functions, attaining an impressive 99.999% availability. This enhancement resulted in a 10x surge in traffic handling capacity, accompanied by cost-saving efficiencies through intelligent descaling during periods of low usage. Tech Stack Used: KEDA, Grpc Server, Go
• Automated CI/CD pipeline using FluxCD for GitOps and AWS CodeBuild, slashing deployment times from hours to minutes. Achieved seamless processes with zero DevOps dependencies. Tech Stack Used: Flux, Helm, Kustomize, AWS
• Developed in-house learning architecture similar to Github Codespaces, Gitpod which can spawn interactable learning environments(pods) like Jupyter Notebooks, Nodejs + Express loaded playgrounds. Tech Stack Used: Go, Kubernetes SDK, gorm, gorilla mux.

Nov 2021 – Feb 2022 · 3 mos · Remote

Jun 2021 – Jul 2021 · 1 mo · Hyderabad, Telangana, India

• Build a centralized configuration store for maintaining uniformity of configurations across all FBA Export Services using Amazon's internal Tree based database, with custom java library for extraction of configuration based on service locations.
• Tech Stack Used: Java, Spring boot, Amazon Web Services, G2S2(Amazon Internal Tree based data structure)

May 2020 – Aug 2020 · 3 mos

• Migrated Prometheus's benchmarking infrastructure tool(Prombench) from Google Kubernetes Engine, GCP to Elastic Kubernetes Service, AWS. Prombench, developed using Kubernetes, employed numerous scrappers and readers with approximately 1,000 pods simultaneously assessing the performance of Prometheus under stress and loads.
• Implemented numerous Grafana dashboards to thoroughly benchmark Prometheus, covering diverse requirements from identifying memory leaks to monitoring buffer consumption per query.
• Tech Stack Used: Golang, Kubernetes, Prometheus, Amazon Web Services, Google Cloud Platform.

May 2019 – Aug 2019 · 3 mos · Troy, New York, United States

• Implemented random input/output generation during evaluation, preventing hardcoded test cases and fostering a more authentic assessment of student learning.
• Engineered a scalable infrastructure, leveraging RabbitMQ queues and a robust worker architecture, to efficiently process student solution evaluations at an exceptional throughput rate—specifically catering to 10,000 students in just one hour and handling a total of 60,000 submissions, during exam hours.
• Tech Stack Used: Python, C++, RabbitMQ, PostgreSQL, Symphony