nodramaops
Get in touch

No-drama DevOps and AI agents that take over the routine

Hi, I'm Stefan, a freelance DevOps engineer. For eight years I've kept production running for product companies, from a game studio I built from scratch to a service handling 45,000 requests a minute. I build reliable infrastructure and calm release processes, AI agents that take over routine work, and turnkey websites.

Message on Telegram See case studies
  • 8+years in production, since 2016
  • 7companies: EdTech, gamedev, high-load, SaaS
  • 45krequests per minute on a high-load product

release v2.42 main → production

  1. Tests214 passed, 48 s
  2. Image build1 m 12 s
  3. Security scan0 critical
  4. Productionapproved by Stefan

Six minutes from merge to production. A person approves the release before production.

Example: a CI/CD release with checks at every step.

receipt.jpg184 KB October, apt 54
Building bot Recognised: ₴1,200, apt 54, generator fund. Added to the sheet, waiting for the money to arrive.
Building bot The money has arrived. Thank you!
What was the money spent on in September?
Building bot September spending was ₴18,400: fuel ₴12,000, generator service ₴6,400. Receipts are in the sheet.

Example: an AI Telegram bot reads receipts and answers questions from its database.

cloud bill, $ per month−40%

  • July$1 240
  • August$1 310
  • Octoberafter the audit$780
  • What we savedper month
  • 3 forgotten servers switched off↓ $210
  • Database on a smaller size↓ $160
  • Logs kept for 30 days, not forever↓ $90
  • Static files via CDN↓ $70
  • Saved per month$530

Example: what a cloud cost audit can find. Figures are illustrative.

  1. monitoring03:12

    FIRINGAPI responds in 1.9 s (threshold 0.8 s)

  2. ops-agent03:12

    Gathered context: the last 15 min of logs and metrics, recent releases. Most likely cause: since v2.41 the database runs out of connections.

  3. ops-agent03:13

    Two options from the runbook. The on-call engineer decides:

    Raise the connection limit Roll back v2.41
  4. Stefan03:15

    Raise the limit. We'll dig into the code during the day.

  5. monitoring03:17

    RESOLVEDAPI is fast again

    0.8 s p95, s 1.9

Example: the AI agent gathers context and suggests options; a human makes the call.

What we can change together

What my work aims at: fewer fires, calmer releases, clearer costs and less routine for your team.

A release takes half a day and someone always stays up late

Releases go through CI/CD with automatic checks and as few manual steps as possible

You learn about outages from your customers

Monitoring and alerts in Telegram or Slack help you spot problems as early as possible

The cloud bill keeps growing and nobody knows why

Spend is broken down by line, waste is switched off, and the provider fits your budget

Only one person knew how it all works, and they've left

Infrastructure is described as code (IaC) with docs, so a newcomer has a place to start

A manager copies requests and payments from chats into a spreadsheet by hand

A bot parses the text, photos and PDFs and keeps the books, while a person checks only the unclear cases

Routine customer questions eat hours every day

An AI agent answers routine questions from your knowledge base and passes the hard ones to a person

Services

Two core tracks that reinforce each other: reliable infrastructure, and AI that takes over repetitive work. Plus turnkey websites.

DevOps and cloud

For teams that need senior-level DevOps without hiring a whole department.

  • Cloud or servers that fit your budgetAWS and Google Cloud, or the more affordable Hetzner, Vultr, DigitalOcean and the like. I pick what fits your load and your wallet.
  • Infrastructure as code (IaC)Servers, networks and permissions are described in code. You can rebuild it, review it and hand it to another team.
  • Kubernetes and GitOpsContainers, autoscaling and test environments to check changes before release.
  • CI/CDFrom commit to production with tests and checks and as few manual steps as possible.
  • Monitoring and alertingMetrics, logs and dashboards. Alerts tuned so you don't drown in noise.
  • SecurityBot and attack protection, single sign-on, secrets kept out of code, private networks and VPN.
  • DatabasesReplication, backups with restore checks, and performance tuning.
  • Infrastructure savingsBill breakdown, right-sized resources and, when it pays off, a move to a cheaper provider.
  • AWS
  • Google Cloud
  • Hetzner
  • Vultr
  • DigitalOcean
  • bare metal

AI agents and bots

For businesses where people spend hours doing the same thing in chats, spreadsheets and email.

  • AI-powered Telegram botsThey understand text, screenshots, photos and PDFs, ask when something is missing, and answer based on your data.
  • AI agents for your workflowsThey triage requests, prepare reports and answer customers. A human approves the important decisions.
  • IntegrationsYour bank, spreadsheets, CRM, email and internal systems.
  • Agent access to your data (MCP)The agent works with your knowledge base, documents and logs within the permissions you grant.
  • AI for DevOpsAn agent reads alerts, logs and metrics, helps find the cause and suggests possible fixes.
  • Low-cost hostingServerless: you pay for real usage, not for an idle server.
  • Claude
  • ChatGPT
  • MCP
  • Telegram
  • Serverless

Turnkey websites

Business-card sites and mid-complexity websites on the stack that suits you: from fast static pages to a site with an admin panel, forms and integrations. Delivered with a domain, HTTPS, hosting and automatic deploys, because infrastructure is my home turf. I built this site too.

  • business-card site
  • landing page
  • admin panel or CMS
  • forms and integrations
  • mobile-ready
Order a website

Not sure where to start? Start with an audit

In 2 working days I review your infrastructure or workflow, look for risks and places to save, and give you a written plan with estimates. The express audit is $100, credited to the follow-up project.

Book an audit

Case studies

Eleven real stories from eight years. No internal company figures, just what changed for the business.

Reliability under load

When every minute of downtime costs money and reputation.

Highload2019 – 2021

45,000 requests a minute across several regions

A product with constant heavy traffic, where even a short outage hits revenue immediately. I led the company's move to containers and Kubernetes, rebuilt releases and brought monitoring of every server in every region into one place.

Servers from every region on one screen, so problems are easier to spot early.

EdTech2024 – now

An online school with live video lessons

Kids and teachers meet in live video lessons every day. I'm responsible for the platform's infrastructure: video servers, databases, cloud, monitoring and incident response.

Changes can be checked in a separate test environment before release.

SaaS2021 – 2023

A design platform for users around the world

A geo-distributed platform running around the clock. I kept core services on call, rewrote the CI/CD pipelines, moved infrastructure to IaC and optimised cloud spend.

Infrastructure described in code, and fewer manual operations for developers.

Launches and migrations

When you need a foundation built, or a move made without breaking anything.

Gamedev2023 – 2024

A game studio: infrastructure from a blank page

The studio shipped mobile games and a Telegram Mini App, but there was no infrastructure to speak of. I designed and built all of it: networks, servers, databases, automatic game builds and backend releases. Every service moved into containers with separate development and production environments, and players sign in through a single account system.

The studio got infrastructure described in code, with monitoring, alerts and cost control.

Product2018 – 2019

A database with a standby in another zone

PostgreSQL ran in a cluster the team had deployed and maintained inside Kubernetes itself. I moved the databases to a managed service with a standby in another availability zone and set up backups of the whole cluster.

If one availability zone fails, the database switches to its standby, and the cluster can be restored from backup.

SaaS2021 – 2023

Builds that scale with the load

Builds and tests ran on a fixed set of servers. I rebuilt the process: build servers spin up for the number of jobs and shut down when there is no work. Along the way I moved the WordPress site into containers and Kubernetes, and the packages from a self-hosted registry to a managed one.

Less waiting for builds at peak hours and fewer servers idling at night.

AI and automation

When the routine can go to a machine and people keep the decisions.

AITelegramOwn product

An AI bookkeeper for a 545-apartment building

Residents chip in for a backup generator and post about it in the group chat. Someone used to spend evenings reconciling it all in a spreadsheet. Now the bot reads the messages, screenshots and PDF receipts, matches bank payments against the statement, asks when something is missing, and answers "what was the money spent on in September?" with figures from its database.

Contributions for 545 apartments tracked with almost no manual entry, and bank payments matched against the statement.

AIEdTech2024 – now

AI assistants for an engineering team

I bring AI agents into engineers' daily work. Through MCP the agents see logs, metrics and the knowledge base, and help investigate problems.

Finding the cause of a problem gets easier: the agent gathers data from different systems itself.

Automation2021 – 2024

Bots that take routine off developers

A code repository bot at a game studio, an automatic library update bot at a product company, and scripts that replaced manual operations.

Part of the routine people used to do by hand now runs automatically.

Security and control

When it matters who has access and who is knocking at your door.

EdTech2024 – now

A platform for people, not bots

I analyse suspicious traffic and build the defences: a web application firewall, rate limits and bot filtering.

More resources go to real users, because the firewall cuts off part of the suspicious traffic at the door.

Highload, Gamedev2019 – 2024

Every permission under control

Moved public servers and databases into a private network, set up central access management and employee onboarding and offboarding.

Access is granted as needed and revoked during offboarding.

How we'll work

From the first call to the handover. You see what is happening and understand what it costs.

  1. Intro call

    30 minutes online. We go through the task and honestly decide whether I'm the right fit. Free of charge.

  2. Plan and estimate

    A written plan: what we do, in what order, how long and how much. Scope and price go into the contract, and scope changes are agreed separately.

  3. Work in stages

    Short stages, each with a result you can try. A short weekly report on what's done.

  4. Handover

    Docs, incident runbooks and every credential are yours. Then you run it yourselves, or I stay on for support.

Pricing

"From" prices depend on scope. I give the exact figure after the intro call and fix it in the contract before we start.

DevOps and cloud

  • Express infrastructure auditA report with risks, quick wins and a work plan. 2 working days. The fee is credited to the follow-up project.

    $100

  • CI/CD for a serviceBuild, tests and deploy with no manual steps.

    from $350

  • Monitoring and alertingMetrics, logs, dashboards, alerts to Telegram or Slack.

    from $550

  • Security and accessBot and attack protection, single sign-on, secrets, backups, VPN.

    from $500

  • Infrastructure savingsA bill breakdown and the changes that bring it down.

    from $400

  • Move to another providerFor example, from an expensive cloud to Hetzner or Vultr, with a migration plan and every step verified.

    from $1000

  • Infrastructure from scratch, or a move to KubernetesIaC, cluster, GitOps, monitoring and docs.

    from $2000

AI agents and bots

  • AI-powered Telegram botAnswers from your knowledge base, request intake, photo and PDF parsing. Low-cost serverless hosting.

    from $500

  • AI agent access to your data (MCP)An agent works with your database, documents or logs, with exactly the permissions you grant.

    from $700

  • AI agent with integrationsCRM, spreadsheets, bank, your systems. A human approves the important decisions.

    from $1500

Websites

  • Business-card websiteOne or a few pages, mobile-ready, with a domain, HTTPS and hosting.

    from $250

  • Mid-complexity websiteSeveral sections, an admin panel or CMS, forms and integrations, on the stack you prefer.

    from $1000

Experience

Started as a server administrator in 2016, and have worked as a DevOps engineer at product companies since 2018.

  1. All Right, DevOps EngineerOnline English school for kids
  2. Vireye Studio, DevOps EngineerMobile games and a Telegram Mini App
  3. VistaCreate, DevOps EngineerGeo-distributed graphic design platform
  4. Together Networks, DevOps EngineerHigh-load product, about 45,000 requests per minute
  5. Onseo, DevOps EngineerOne cluster for development, QA and production
  6. Smile Expo, DevOps EngineerAbout 200 servers on in-house virtualisation
  7. Pay Per Call, Linux administratorAbout 450 domains, mail, VPN, backups

MSc, Telecommunications and Radio Engineering, State University of Telecommunications, 2021

BSc, Telecommunications, National Aviation University, 2017

Languages: Ukrainian native, English professional, German elementary

What I do

Clouds and providers
  • AWS
  • Google Cloud
  • Hetzner
  • Vultr
  • DigitalOcean
  • Gcore
  • bare metal
Infrastructure
  • IaC
  • Kubernetes
  • containers
  • GitOps
  • networking and DNS
Delivery
  • CI/CD
  • a test environment per change
  • automatic dependency updates
Observability
  • metrics
  • logs
  • traces
  • alerting
  • on-call
Data
  • SQL and NoSQL databases
  • queues
  • replication
  • backups
Security
  • WAF
  • bot protection
  • SSO
  • secrets management
  • VPN
AI
  • AI agents
  • MCP
  • Telegram bots
  • Claude
  • ChatGPT
Websites
  • business-card sites
  • landing pages
  • sites with an admin panel
Video and realtime
  • WebRTC servers

FAQ

Do you work with small teams and startups?

Yes. That's often where DevOps is needed most, while a full-time hire is still too early.

Does it have to be AWS or Google Cloud?

No. The big clouds are convenient but not always necessary. For many projects Hetzner, Vultr, DigitalOcean or your own servers often deliver the reliability you need for less. I'll help you choose and, if needed, move.

What do you build websites with?

On whatever stack is easy for you to maintain: from simple static pages to popular frameworks and CMSs with an admin panel. If you have no preference, I'll suggest the simplest option that loads fast and is cheap to host.

Can we start with a small task?

Yes. An audit, an alerting setup or a single bot are good starts. You see how I work without a big commitment.

How much does an AI bot cost?

From $500 for a bot that answers from your knowledge base and takes requests, and from $1500 for an agent integrated with a CRM, spreadsheets or a bank. I give the exact figure after the intro call. Serverless hosting is cheap for most bots because you pay only for usage. AI model requests are billed separately as you use them.

Who owns the code and the access?

You do. Repositories, cloud accounts and secrets are yours. I work with the permissions you grant and hand everything over at the end.

How is the work formalised?

Officially, through my Ukrainian sole proprietorship (FOP): a contract, an invoice and a certificate of completed work. Payment in UAH to the FOP account, in stages.

Do you sign NDAs?

Yes, no problem.

Tell me about your project

Telegram is the fastest way to reach me. The first call is free.

Message on Telegram

Or leave a request

Preferred way to talk
What you need

The request goes straight to my Telegram. Only I see your contact details.