RU/EN
Message ↗

Headphones recommended

Valery — AI specialist: bringing AI agents into business processes

#NOTARESUME — a list of shipped tasks

Eight autonomous systems. Zero employees, one AI subscription, zero promises. While you're reading this line, they keep working.

All of this costs 2,000 ₽ a month. The proof is below.

  • 4,000+users
  • 12games
  • 4serversworkshop, payment processing + hosting, gateway, proxy server
WHO'S AT THE CONSOLE

Valery

AI specialist: I turn routine work into AI automation that runs unattended

I bring AI agents (Claude Code) into business processes — from the task to a working system with tests, monitoring and human control.

Nizhny Novgorod / Yerevan / Moscow · remote

  • ~2,000hours of hands-on work with AI agents
  • 20+systems in operation
  • 500+days without a crash for one bot

BRING ME

  • Reports and data from several systems are put together by hand.
  • A stream of requests, leads, CVs or documents needs sorting and filtering.
  • You need a bot or an AI assistant for staff or customers.
  • Your team wants to work with AI agents but doesn't know how to organise it.

HOW I WORK

  1. I take the process apart and find where automation pays off.
  2. I give the task to AI agents and accept the result only by checks: tests, a live run.
  3. I launch it and keep it running: logs, backups, monitoring, alerts. Code does the maths, AI only words it.

Linda — AI job screening that repairs itself

Every 15 minutes it collects vacancies from hh.ru and Habr Career; one model drops what's clearly off, a second scores the rest against the profile: of 7,300+ vacancies, under 3% reached a human. Applying happens only after their decision. When a site changes its markup, a watchdog spots the breakage and AI fixes the parser in a sandbox; the fix is accepted only if all 134 tests pass.

For a business, that's monitoring and qualifying leads, requests or tenders with a human in control.

I don't write code by hand: I run AI agents and answer for the result. Instead of years on payroll — working systems you can open and check below.

SYSTEM 01

Games by the “Friday” studio, plus promo materials

The full cycle — from clean, fast code to promo videos. While a competitor is still loading megabytes of framework, our game has already taken the player's first move.

A separate system first develops a game, then puts it in the maintenance queue. It prepares promo materials and store-page copy for the platforms. Scripts launch the game in a headless browser on their own, record gameplay in vertical and horizontal formats, check that no Cyrillic has leaked into the English build, and cut the video. The full cycle up to revenue on the platform: platform SDKs, ads, in-game purchases, VK payments, cloud saves, leaderboards, passing moderation. The terms of use and the privacy policy for every game are drawn up in-house. Nine games in three months on JavaScript, Three.js, Godot and Unity — one of them is being prepared for Steam, with around 1,500 automated tests and checks across the projects.

The studio's game pipeline
  1. 3concept
  2. 1in development
  3. 4in review
  4. 4live on VK
  • Slovokhod

    Published on VK. A hundred word-search levels and not a single audio file in the build: the game synthesises every sound itself.

  • Cat Nonograms

    Published on VK. 130 nonograms across themed chapters, with daily puzzles.

  • Color and Shape Sort

    Published on VK. Sort pieces by both colour and shape, so colour is not the only cue.

  • Two Fortresses

    Published on VK. Recruit a squad, upgrade your economy and control a hero on a single battlefield.

HOW THE STUDIO IS BUILT

Four servers, each with its own job

The workshop where games are made is kept apart from player data, and a single gateway meets everything from outside. Addresses and internals aren't published: not a trade secret, just basic hygiene.

players and visitors
Gatewayserver in Moscow

The single point of entry. Accepts every incoming connection and routes each one to the right department.

Workshopdevelopment server

Where the work happens: agents write code, make builds, run automated tests and publish test versions.

Payments and hostingrelease server

Payment processing, player data storage and game hosting. Kept apart from the workshop: development never touches what belongs to players.

Proxy servernetwork route

The backup route: if a data centre fails, the studio's traffic switches over to it automatically.

STUDIO SERVICES

  • Payment processingrunning

    In-game purchases: they are processed and stored on our own server.

  • Leaderboardsrunning

    Player high-score tables on our own server.

  • Analyticsrunning

    An independent data collection system: 100+ tracking points per project. And that's on top of each platform's own analytics — VK, Yandex and others.

  • Legal documentsrunning

    The terms of use and the privacy policy for every game are drawn up in-house.

  • Contentrunning

    Two channels built on photos and text. Plus two video lines — each published automatically on two or more platforms.

SYSTEM 02

Human health comes before any automation

While you're in the flow and losing track of time, the board keeps an eye on what you breathe. CO₂ and dust are invisible, but they drain your HP. Your nose can't sense them — the board can. (Actually, the board doesn't care. We taught it to worry.)

A board with seven sensors sends temperature, humidity, pressure, light level, volatile compounds, three fractions of dust and CO₂ to the server once a minute. When the server is unreachable, the board notices by itself, buffers readings in its own memory and sends them with their real timestamps as soon as the connection is back. One of the sensors turned out to be a clone with its own protocol: we decoded it by hand, standard libraries can't see it. The setup moves to a greenhouse, a server room or any workplace without a single change to the logic.
SYSTEM 03

Media. What's the use of all of the above if nobody knows about you?

I filled a folder with 169,000+ photos once — from then on the channel runs itself: it posts during the day at unpredictable times, so the feed doesn't look robotic.

Two bots. The first posts a set of nine photos with a caption once a day and moves what it used into an archive, so nothing repeats. The second wakes up on schedule: text at 5:50, video at 6:30; it collects material in advance from source channels by itself and hands it out one piece a day. Both have an admin panel right in the chat: status, manual run, subscriber analytics, automatic database backup. The first has run as a system service for months without a crash. The second was started by hand and still hasn't missed a single post.
SYSTEM 04

A fix for the RuNet that's coming.

The bridge connects SOCKS5 to HTTP and brings Telegram Desktop back to working order in a couple of minutes.

A local server starts on the computer and takes over all the dirty work: authorisation, tunnelling, handling connections, with a log of every one of them. The instructions are built like proper diagnostics, from cheap to expensive: first two checkboxes in the app's own settings, and only if that didn't help does the bridge start. It runs as a system service, self-test green. The task is a standard one: the same cure works for any application that speaks one protocol while you have another one paid for and available.
SYSTEM 05

All in

what?

Crypto portfolio analytics that sees where the market punishes the greedy: the market situation, your assets, a comparison of instruments and the maths of risk. It doesn't press “Buy” itself — a trading crypto bot for the exchange built on top of it is in development now.

The system reads a portfolio on the exchange and explains in plain language what is happening to it: the exact picture of the money, concentration, drawdown, liquidity, a calendar of token unlocks. It sees when the market situation has changed and there is potential to earn more by moving into another asset or changing the way it yields. It gives no trading signals and doesn't predict prices — by a deliberate decision. All numbers are calculated by Python using carefully verified formulas in a separate module — the module runs on my server and updates the calculations on a schedule; the language model receives ready-made figures and writes the sentences, so no invented numbers appear in the report. The portfolio on the Russian page is the same data structure filled with generated figures: real amounts and holdings aren't published for privacy reasons, not because the connection isn't ready. The framework carries over to any operational monitoring where the conclusions are worded by AI. The next stage is a trading crypto bot that will work on the exchange by itself: it's in development.
SYSTEM 06

Our own cartoons

One production line serves three channels: the script, the voice-over and all the material for editing are assembled and numbered — the editor only has to create.

The system gets a topic and carries on by itself: writes the story, splits it into scenes, sets up characters and settings, lays it out shot by shot, assembles image descriptions, generates images, records the voice-over, edits and exports almost-finished material — all that's left is to put the video together in Premiere. The story is written once in canonical form; the English, Spanish and Russian versions are derived from it. Twelve videos have already been assembled from start to finish in a technical run — still with placeholder pictures: image generation is connected, the visual style is still being chosen. All generation runs on an ordinary home computer, with no subscriptions or APIs — including the review stage, where an LLM itself checks that the script follows the plan, the characters in the pictures don't change, and the overall setting holds. This way several cartoon series can run in parallel.
SYSTEM 07

An operations HQ of eight autonomous agents

Every project needs focus. To keep it from blurring as I switch between tasks, I set up a department that has effectively become my deputies. I compared five methodologies for working with agents on one task and kept the one with the best result.

A shared chat where eight different characters live: each has a name, a manner of speaking and its own memory area. No noise, only pragmatic calculation. Sometimes they joke around — I couldn't do without that option. No project here is frozen: all eight keep developing, and the agents are responsible precisely for that development. Today the HQ lives in Telegram; soon it will run in MAX in parallel.
SYSTEM 08

AI makes mistakes even in the right hands. It needs an oversight body.

Every 20 minutes an autonomous system checks the network for data leaks and unauthorised intrusion. It is forbidden to act on its own — only with the founder's permission.

A full check of the local network for internal and external threats: from the outside, a public IP leads straight to the router, and behind it stand four equal devices — a server, a PC, a nettop and a laptop. For months the server's network share exposed the entire home directory for reading, and it held an exchange key and bot tokens. One computer served its user profile anonymously, without a password. The report separately records a case where drafts of the audit itself once saved some of the secrets in plain text to that same shared folder: the inconvenient fact stayed in the document. The most useful thing in the work turned out to be the order of actions. First close the doors, then rotate the secrets — and do it from a device known to be clean.

Every system above has a live, interactive version — widgets, not just text — on the Russian page.

Infrastructure upkeep
Rented server300 ₽
Third-party proxy200 ₽
AI subscription1500 ₽
Total per month2,000 ₽

Infrastructure upkeep: 2,000 roubles a month

The line-by-line breakdown is in the block next to this. Nothing else was needed.

I build systems, and sometimes I can get too deep into them. So that I don't end up at the clinic later, HP100 keeps an eye on my room. And keeping all of this running takes 2,000 roubles a month.

I take on tasks that are interesting to solve: AI-agent automation, bots, monitoring, games. Write to me with what needs doing — I'll reply and tell you how I'd approach it. If I take it on, I'll take your problem off your hands and give you back a working mechanism.

Request

Describe a routine task in two sentences — I'll tell you how to automate it and how long it would take.

0/3000 characters

I'll read your request myself. To break it down into points, the text is processed by an AI (Google Gemini); I don't share it with anyone else.