Senior DevOps / SRE Engineer
5+ Years · Kathmandu, Nepal · open to remote
building ai agents for ops · open to remote senior roles
Swedish multi-tenant event-management SaaS. Sole platform owner. ECS Fargate services behind ALB, Aurora PostgreSQL with schema-per-tenant, ElastiCache Redis, Amazon MQ. Tenant routing via DynamoDB registry. Customers across Europe.
ECS Fargate · Aurora PostgreSQL · ElastiCache · Amazon MQ · DynamoDB · CloudFront
Cross-region disaster recovery from eu-north-1 to eu-west-1. Built where none existed. Aurora Global Database for sub-second cross-region replication, EFS and ECR replication, shared KMS keys across regions. Documented runbook for promotion.
Aurora Global DB · EFS · ECR · KMS · Route 53
Python tenant-provisioning orchestrator. One API call sets up schema-per-tenant on Aurora, wires SQS and EventBridge, creates ALB listener rules, provisions a CloudFront / S3 distribution, configures Route 53 records, and registers the tenant in DynamoDB.
Python · FastAPI · Aurora · SQS · EventBridge · Route 53
Diagnosed a 19-minute full-platform outage caused by blocking Redis KEYS calls exhausting the Tomcat/JDBC thread pool. Added connection-pool checkout timeouts, tuned RDS parameters, and drove a 68-task reliability program across 11 epics and 7 sprints to prevent recurrence.
Aurora · Redis · JDBC · Postmortem · SLOs
Self-hosted OneUptime on K3s in eu-central-1 (separate region from primary). Status pages, uptime monitoring, on-call scheduling, incident management. Designed so observability survives a primary-region outage.
K3s · OneUptime · OpenTelemetry · Loki
OpenTelemetry collector dual-exports metrics, logs, and traces to OneUptime and Loki at the same time. Consolidated fragmented monitoring into one observability stack. Grafana sits on top.
OpenTelemetry · Loki · Grafana · Prometheus
Self-managed three-node Elasticsearch cluster managed with Terraform and Ansible. Split deploy and split-restart playbooks so a single config change cannot cascade across the cluster.
Elasticsearch · Terraform · Ansible
CI/CD pipelines spanning Jenkins, GitLab CI, and AWS CodePipeline / CodeBuild. Targets include ECS, Lambda, CloudFront, and EC2 deployments. App and infra share pipeline patterns.
Jenkins · GitLab CI · CodePipeline · CodeBuild · Docker
Cut monthly AWS spend by removing orphaned NAT Gateways, adding S3 and DynamoDB gateway endpoints to drop data-transfer cost, setting log-retention policies on CloudWatch, and right-sizing EBS volumes.
VPC Endpoints · CloudWatch Logs · EBS · NAT
Open-source Model Context Protocol server that wires AI assistants like Claude into the Hashnode content API. The pattern carries into the broader agentic-DevOps work. (Note: Hashnode has since wound down public API access.)
Python · MCP · Hashnode API
github.com/sbmagar13/hashnode-mcp-serverMultimodal text-to-image generation using VQGAN + CLIP architectures in PyTorch. From the AI/ML era of the career, kept here as an artifact.
PyTorch · CLIP · VQGAN · Python
github.com/sbmagar13/VQGAN-CLIP-Text-to-ImageKubernetes (K3s)
Stood up OneUptime on a K3s cluster in a separate region (eu-central-1) for status pages, uptime monitoring, on-call scheduling and incident management. Deliberately off the primary region so observability survives a primary-region outage.
Docker
Containerised build and deploy pipelines on Jenkins, GitLab CI, and AWS CodePipeline / CodeBuild. Targets include ECS, Lambda, CloudFront and EC2. App and infra share pipeline patterns so a single change can flow through any of them.
AWS
Diagnosed and resolved a 19-minute full-platform outage caused by blocking Redis KEYS calls exhausting the Tomcat/JDBC thread pool. Added connection-pool checkout timeouts, tuned RDS parameters, then drove a 68-task reliability program across 11 epics and 7 sprints to prevent recurrence.
Aurora PostgreSQL
Established a cross-region DR path where none existed: Aurora Global Database from eu-north-1 to eu-west-1, EFS and ECR replication, shared KMS keys, and a documented runbook for promotion.
Terraform
Self-managed three-node Elasticsearch cluster orchestrated with Terraform and Ansible. Split deploy and split-restart playbooks so a single config change cannot cascade across the cluster.
Python
One Python API call sets up schema-per-tenant on Aurora, wires SQS and EventBridge, creates ALB listener rules, provisions a CloudFront / S3 distribution, configures Route 53 records, and registers the tenant in DynamoDB.
Grafana
Consolidated fragmented monitoring into one stack. Grafana over Prometheus, Loki and CloudWatch with per-cluster, per-namespace and per-tenant dashboards. Alert routing wired to OneUptime on-call.
OpenTelemetry
OpenTelemetry collector dual-exports metrics, logs and traces to OneUptime and Loki at the same time. The duplication is the point: if a primary-region failure takes the main observability stack down, the OneUptime side still pages.
Anthropic MCP
Self-learning track. Building MCP-based agents that wrap real DevOps tasks (log triage, runbook prompts, infra analysis) so Claude and similar assistants can drive them. Built an open-source Hashnode MCP server that wires AI assistants into the Hashnode content API. Current focus is the broader agentic-DevOps stack: MCP, LangGraph, local LLM inference.
VolgAI, Genese Cloud Academy, IBZ Networks. Built AI chatbots with RASA (NLP), backend APIs with Django and Flask, RTSP/FFmpeg pipelines for CCTV image processing. Async work via Celery and RabbitMQ. AWS AI/ML Internship at Genese.
Graduated Bachelor in Computer Engineering from Tribhuvan University (IOE, Pokhara). Joined Cloudyfox Technology in September as DevOps Engineer. First Terraform / Terragrunt at scale across AWS.
Ran Kubernetes for containerized workloads at Cloudyfox. Built CI/CD on GitLab CI and Jenkins for both app and infra deploys. SysOps, Linux admin, OpenVPN, centralized logging with CloudWatch, ELK, Grafana.
Joined Threadcode Technologies as DevOps / SRE for EventLogic, a Swedish multi-tenant event-management SaaS. Owner of the entire AWS platform end to end: ECS Fargate, Aurora PostgreSQL, ElastiCache Redis, Amazon MQ in eu-north-1. Technical Reviewer for Python for DevOps (Packt).
Diagnosed and resolved a 19-minute platform outage caused by blocking Redis KEYS calls exhausting the JDBC thread pool. Drove a 68-task reliability program across 11 epics and 7 sprints. Built cross-region DR from eu-north-1 to eu-west-1 with Aurora Global Database, EFS replication, and shared KMS keys.
Deployed self-hosted OneUptime on K3s in eu-central-1 for status pages, uptime monitoring, and on-call. OpenTelemetry collector dual-exports to OneUptime and Loki so observability survives a primary-region outage. Shipped an open-source Hashnode MCP server that wires AI assistants like Claude into the Hashnode content API.
Still running the production platform end to end. Hardening the multi-region story, sharpening SLOs, and going deep on AI agents for DevOps work: MCP, LangGraph, local LLM inference, self-learning side projects that wrap real ops tasks. Shipped this 3D portfolio as the public face of the practice. Open to remote senior DevOps / SRE roles.
Route 53 (4y), VPC Networking (4y), Terraform (4y), Terragrunt (3y), CloudFormation (3y), AWS CDK (2y), Ansible (4y), Docker (4y), Kubernetes (K3s) (3y)
GitLab CI (4y), Jenkins (4y), CodePipeline (3y), GitHub Actions (3y)
AWS (4y), ECS Fargate (3y), AWS Lambda (4y), Amazon MQ (2y), CloudFront (4y), S3 (4y), API Gateway (3y), EFS (2y), ECR (4y)
Aurora PostgreSQL (3y), ElastiCache (Redis) (3y), DynamoDB (3y), AWS DMS (2y), PostgreSQL (4y), Redis (3y)
Prometheus (3y), Grafana (3y), Loki (2y), OpenTelemetry (2y), AWS CloudWatch (4y), OneUptime (1y), ELK Stack (4y), Elasticsearch (4y)
Python (5y), FastAPI (3y), Django (3y), Flask (3y), Bash (5y), JavaScript (4y)
AWS IAM (4y), AWS KMS (4y), Secrets Manager (3y), AWS Inspector (2y), CloudTrail (4y), OpenVPN (3y)
Anthropic MCP (1y), LangGraph (1y), LangChain (1y), Local LLM Inference (1y), PyTorch (2y), RASA (2y)
Arch Linux (5y), Ubuntu (6y)
Apache Airflow (2y), Airbyte (1y), Celery + RabbitMQ (3y), FFmpeg (2y)