Saumitra Patil

•Maintained and optimized Kubernetes clusters and VMs, reducing infrastructure cost by ~$34K/month through, unused resource cleanup, and node pool right-sizing, maintaining 98–100% uptime. •Led migration from Google Kubernetes Engine to Amazon EKS, re-architecting workloads and designing custom VPCs, IAM roles, GPU node groups, and AWS CodePipeline/CodeBuild CI/CD pipelines. •Built full observability stack using Prometheus, Loki, Promtail, Tempo, and Grafana Alloy, including GPU monitoring via NVIDIA DCGM on EKS. •Drove SOC 2 Type 2 readiness implemented KMS-based secrets encryption, enforced least-privilege IAM with group-based access, configured audit logging, and hardened firewall rules. •Architected secure, keyless access to EC2 instances using AWS Systems Manager (SSM) with IAM Identity Center, and set up rate limiting at NGINX and Ingress levels. •Provisioned and managed LLM serving infrastructure using vLLM and Ollama on GPU VMs and Kubernetes, including model caching and NVIDIA device plugin configuration. •Deployed self-hosted internal tools (n8n, Listmonk, other internal tools) using Docker Compose on VMs, automating newsletter workflows and internal operations. •Mentored junior engineers on Kubernetes, cloud architecture, and DevOps best practices.

I lead all cloud and infrastructure efforts at Bridge-it. - Fully managed Google Cloud Platform (GCP) environments across development, staging, and production, ensuring proper infrastructure operations and deployments. - Led weekly production releases, incorporating automated post-deployment tests, with slack notification integration for each. - Architected and implemented an event-driven Pub/Sub system using Google Pub/Sub, creating custom topics and subscriptions for real-time message delivery to both the Copilot system and a Django-based Notification service. - Configured and tuned PgBouncer as a connection pooling layer between Google Cloud Run services and Cloud SQL (PostgreSQL), mitigating connection exhaustion during serverless scaling and improving overall database latency. -Provisioned secure VNC access to macOS-based EC2 instances, enabling GUI-based remote access for the testing team by configuring Remote Management.