Galust Arakelyan

IT Smart Finance — fintech lending group (JoyMoney, Boostr, Nibble)

• Backed up production after the cloud provider lost two of three availability zones and worker nodes had no internet access: ran Velero outside the cluster to capture the full 349-workload cluster (10,466 objects, 0 errors), saved all 123 production images (19.7 GB after layer dedup) and Vault, Redis, RabbitMQ and Redpanda data, all checksum-verified.
• Restored production on new on-prem Kubernetes (Talos, Ceph) within one day: Vault from a Raft snapshot, Redis (425K keys), RabbitMQ (781 queues, 5.9M messages), mTLS proxies to credit bureaus and banks, and 125 services staged safely at zero replicas.
• Prepared the production move to VMware Cloud Director: sized the platform from real peak usage (10 VMs, 100 vCPU, 328 GB RAM), built a 3-node Vault Raft cluster, RabbitMQ, Redpanda with SCRAM, ACLs and mTLS, and Redis, and staged 273 deployments and 347 ingresses through a Velero pipeline.
• Migrated Vault from file storage to Raft: moved 47.6K keys with 13.7 minutes of downtime and zero drift, added scheduled snapshots to S3 and verified the chain end to end, restore included.
• Built a new production environment for a credit-bureau reporting product in 1.5 days: migrated 3 PostgreSQL databases (54 GB, 351M rows, zero mismatches), set up Kubernetes 1.36 on Talos, Vault, RabbitMQ and GOST mTLS to credit bureaus.
• Moved 16 stateful components (Redis, Redis Cluster, RabbitMQ) in 4 environments to new image sources after the Bitnami catalog change, 7 of them in production night windows, with disk snapshots before every step and no data loss.
• Rolled out logging on Loki and Grafana Alloy with MinIO storage and 30-day retention, then cut its resource budget 4x after load measurements.
• Root-caused about 50 production and staging incidents from evidence: apiserver audit logs, Loki, OpenSearch offsets, Helm history. Examples: a Redis Sentinel OOM caused by chart drift (192 MiB in Git vs 4 GiB live), Terraform state 503s from NO_PROXY in the runner config, and 12,903 log records hidden by lost trace fields, recovered with OpenSearch ingest pipelines.
• Maintained shared GitLab CI templates and Terraform pipelines: fixed the shared DB-migration job, restored Terraform CI in 6+ projects and added release and manual deploy jobs for 3 isolated QA environments (Istio Ambient, Vault policies via Terraform).
• Automated bulk changes with dry-run, check-and-set writes and rollback logs: updated 246 Vault secrets with 0 errors and fixed post-migration 403s on 132 of 136 ingresses.

Stack: Kubernetes (Managed, Talos), Yandex Cloud, VMware Cloud Director, Terraform, GitLab CI, Helm, werf, Vault, Velero, PostgreSQL, RabbitMQ, Redpanda, Redis, Loki, Grafana, MinIO, Python

Sber — banking and technology group

• Built and maintained Jenkins pipelines that build, test and deploy services and ML models to Kubernetes.
• Delivered workloads through Argo CD and Helm (GitOps), keeping environments consistent from test to production.
• Set up Prometheus metrics and alerting for platform and model-serving services; worked with data scientists on packaging and shipping models.

Stack: Kubernetes, Jenkins, Argo CD, Helm, Prometheus, Grafana, Docker

MegaFon — mobile network operator

• Administered a fleet of about 500 Linux servers, including DNS recursors, for the operator's internal infrastructure.
• Operated authoritative DNS on BIND (1 primary, 3 secondaries) with BGP routing (BIRD) and NTP.
• Deployed and supported the in-house DNS self-service app: admins and engineers manage records and zones through it instead of editing servers by hand.
• Automated server configuration with Ansible; maintained Zabbix and Grafana monitoring for DNS and platform services.

Stack: Linux, BIND, BIRD, NTP, Ansible, Zabbix, Grafana, Bash

Sheremetyevo International Airport

• Supported flight schedule and resource planning systems running on the airport's on-prem Kubernetes.
• Upgraded the cluster from Kubernetes 1.23 to 1.29 one minor version at a time, following the version skew policy: etcd snapshot before each step, control plane first, then workers drained and upgraded node by node.
• Before each step, migrated manifests and Helm charts off the APIs removed between 1.25 and 1.29, so workloads kept deploying after every upgrade.

Stack: Kubernetes, Helm, Docker, Linux

Vnukovo International Airport

• Administered Linux servers for internal airport systems.
• Containerized in-house schedule-related applications with Docker, then moved them to an on-prem Kubernetes cluster.

ProstoDevops — online DevOps school

• Oversee about 100 students in two cohorts as senior curator; my own group of 10 closed 59 of 60 technical defenses, the best result among groups, and 3 students were hired during the course.
• Created 20 lesson decks (487 slides) and hands-on labs: Linux, networking, Docker, Kubernetes (kubeadm cluster from scratch, RBAC, etcd backup, upgrades), Terraform, Ansible, Argo CD.
• Wrote the curator handbook; run weekly curator syncs, 1:1 reviews and mock interviews.

• Built and run a commercial multi-node VPN service: a master and 3 nodes on VLESS + Reality (3x-ui), about 50 paying users and 25 online daily.
• Wrote Python services for payments, subscription links in 5 client formats with split routing, and node monitoring with Telegram alerts; Nginx, certbot and systemd on every node.