Arthur Perchatkin

Cloud · DevOps · Linux · AWS · Azure · IaC

I take source code to running infrastructure — then keep it observable, secure, repeatable, and cheap to run.

Pacific Northwest · open to cloud, DevOps, platform, and Linux roles

---

Linux-first. I work the seam where infrastructure, application code, automation, and operations meet.

Layer	What I use
Provision	Terraform · AWS · Azure · VPC/VNet · EC2 · ECS Fargate · RDS · Lambda · API Gateway
Configure	Ansible · Bash · PowerShell · cloud-init · Linux services · Active Directory
Build	Python · TypeScript · React · Flask · FastAPI
Ship	Docker · ECR · GitHub Actions · Nginx · ALB · autoscaling
Secure	IAM · security groups · CloudTrail · Trivy · Checkov · RBAC
Observe	CloudWatch · logs · alarms · NVML telemetry · SQLite findings
Cost	LocalStack · inventory · budgets · teardown · remote state

Selected builds

Project	What it proves
AWS + Flask	Terraform VPC, public/private nets, EC2, ALB, autoscaling, IAM, S3 state, TLS, Nginx/Gunicorn, GitHub Actions
GPU validation + onboarding gate	NVML + CUDA burn on-node; Lambda/API Gateway/DynamoDB decide PROVISION / HOLD / RMA off-node
AWS security auditor	Boto3 checks across IAM, EC2, S3, SG, RDS, CloudTrail → JSON/HTML, Docker
E-commerce analytics	FastAPI on ECS Fargate, RDS, ALB, Terraform, Trivy, Checkov
AD on Azure	VNet/NSG, Windows Server 2022, AD DS, GPO, PowerShell identities, documented lab spend
Gridwatch	Live network rules → SQLite findings → SNS/webhook alerts

Lab index · All public repos

GPU node validation — one control-plane example

The node reports facts. The gate decides. A machine never approves itself.

NVML health ─┐
CUDA burn  ──┼─> combined report ─> API Gateway ─> Lambda ─> PROVISION
Network    ──┘                                      │       HOLD
                                                    └─────> RMA

• Idle PCIe downshifts — treat static link as conservative, prove it under CUDA load.
• ECC is capability-aware (N/A on consumer silicon, counters on datacenter cards).
• One Ansible playbook: health → burn → network → reconcile → submit.

How I work

01  Reproduce the failure before guessing.
02  Read logs and telemetry before changing the system.
03  Automate repetition, not understanding.
04  Keep infrastructure reviewable and disposable.
05  Security, observability, and cost are design — not afterthoughts.

Build · Secure · Observe · Automate