Pavol Odlevak

Hi, I'm Pavol, a DevOps and platform engineer. I have spent 15+ years running production infrastructure across cloud, hybrid, and on-prem, fully remote from Martin, Slovakia since 2013, working across EU, UK, and US time zones.

I started on physical servers. At adlodge, a German ad-tech company, I built and racked them myself and administered them across Frankfurt, Vienna, and an office in Žilina. Novosense came next, where I led the move of its tracking system onto AWS, with Puppet managing the configuration. Then Demotix, a photojournalism platform: I was its only DevOps engineer, running dedicated VMs at a small cloud provider under configuration management, with automatic failover between the cache and load-balancer nodes.

In 2016 our team moved to BENlabs, a Gates Ventures company, where I took the platform in-house from a contractor and, as DevOps Lead, grew the AWS estate from one account into a governed, multi-account organization. As Director of DevOps I led a team of four supporting about a hundred accounts across all of engineering.

When a restructure in 2023 dissolved that team, I stayed on as the sole DevOps engineer and rebuilt the product-placement platform from scratch: seven services on ECS and Fargate, everything as code, cost and security designed in from the start, running at about $3K a month. Along the way I consolidated the estate from a hundred accounts to thirty. BENlabs closed in April 2026, after a 37-year run.

One account, to a hundred, to thirty. I have grown an estate, governed it with a team, and run it alone and lean. Those are the jobs a growing team needs done.

Today I do that for other teams: fractional platform and infrastructure leadership and hands-on engineering across cloud and on-prem, for companies that need senior infrastructure work without a full-time hire. The work is B2B, invoiced through my Slovak company.

Since 2025 I have run agentic AI workflows as part of the daily job, using Claude Code to write infrastructure code, CI/CD pipelines, and runbooks, and to review pull requests. The safeguards matter more than the speed: every change the agents produce goes through the same pull-request review, security scanning, and sign-off as anything a person writes, so nothing rests on trusting the model's output. I run them in real work, and I do not sell a course on them.