Engineering manager, engineer and mentor

18+ years in IT. Built departments from scratch and ran SRE and DevOps teams of up to 15. I build systems end-to-end together with AI agents and set up how teams work with them. Co-author of the SENAR, RENAR and PHASE standards.

18+ years in IT Teams up to 15 30+ projects 3 standards
Contact on Telegram View projects

What people come to me with

So you can tell whether this is your case — not just what I can do

Size
From a founder with no team to a company of 250 — small and mid-sized business
Industries
Healthcare and lab services · EdTech, media and content · Retail, product labelling, logistics · Legal services, real estate, cadastre
Situation
No product yet · The product has hit a ceiling · Legacy nobody dares touch · A team adopting AI and drowning

As an engineering manager

  • —Build a department from scratch: hire, set up process, get it running
  • —The people are there, the results are not — process, releases, ownership
  • —Infrastructure audit: look at it, say honestly what it is, make it reproducible
  • —Take technical leadership for a while — cover a gap or carry a transition

As an engineer

  • —You need the whole product and have no team — architecture through deployment
  • —You have a team, but one block is beyond it: AI, Rust, infrastructure
  • —The system works and nobody knows how — it needs analysis and documentation
  • —People do by hand what a program should be doing

As a mentor

  • —The team already uses AI agents, but the output is unpredictable and unverifiable
  • —Mentoring specific engineers rather than the process as a whole
  • —Development process audit: where exactly it leaks
  • —Adopt SENAR or RENAR as a formal standard

What I don't take on

Projects with no direct access to the people who make decisions. Through three layers of sign-off I cannot be accountable for the result — and I want to be.

What I do

Turnkey SaaS

I build the entire product: backend, frontend, billing, monitoring, deployment. You get a working service, not a set of components.

Legacy Audit

I analyze existing codebases, write technical documentation, migrate infrastructure to IaC. New developers onboard in days, not months.

Custom Development

Web apps, desktop utilities, workflow automation. I work with complex domains and build products that non-technical users can understand.

Consulting and mentoring

I set teams up to work with AI agents: where an agent speeds you up, where it lies, and how to verify the result. I review architecture and help you commit to a decision.

AI/ML Integrations

Embedding AI into business processes: model fine-tuning, RAG systems, multimodal pipelines. From prototype to production with real metrics.

Infrastructure & Architecture

Designing infrastructure from scratch, auditing existing setups. CI/CD, monitoring, IaC, hybrid clouds. Building platforms that scale and don't depend on a single person.

Standards I authored

Three normative documents on engineering in the age of AI — from development methodology to enterprise architecture

SENAR

Co-author — the first AI-native development methodology

Supervised Engineering & Normative AI Regulation. A formalized approach to managing AI agents in software development: roles, quality gates, metrics, knowledge capture.

senar.tech
RENAR

Co-author — requirements engineering for AI-native development

Requirements Engineering and Normative Adaptive Regulation. Requirements, specifications, test cases and adaptation artifacts for projects where AI agents write the implementation. Works standalone, interoperable with SENAR.

renar.tech
PHASE

Co-author — the architecture of the cyber-enterprise in the age of AI

The enterprise as an automatic cybernetic system: all application logic is expressed with three primitives — phase, spawning, subscription. A human and an AI agent change an object's state through the same mechanism, and the degree of autonomy is a setting rather than an architectural rewrite.

phaseconcept.tech
TAUSIK GitHub

A standard taken all the way to executable code

SENAR and RENAR implemented as a working tool: an agent cannot edit files without an open task, nor close one without evidence. 1197 tasks closed, 7115 tests, 76% coverage, Apache 2.0. This site was built under it.

Python 3.11 SQLite + FTS5 MCP ed25519

Projects

Selected examples — from SaaS to system software

Sortula — Smart Bookmarks

sortula.ru

SaaS for semantic bookmark search. From idea to production in 6 weeks: auth, AI categorization, vector search, billing — one developer.

FastAPI Next.js 15 PostgreSQL pgvector

YKademia — EdTech Platform

ykademia.ru

EdTech SaaS in 8 weeks: AI tutor, learning material generation, progress tracking. Self-hosted LLM — zero external API costs.

SvelteKit FastAPI Ollama PostgreSQL

B2B SaaS for LLM Brand Monitoring

Brands don't know how ChatGPT, Gemini, and DeepSeek talk about them. Connected 8 LLM APIs, automated data collection and analytics, billing via CloudPayments.

FastAPI Next.js 15 PostgreSQL Celery RabbitMQ

Legacy System Technical Passport (C#)

A .NET 4.0 system evolved over 12 years with no documentation. Created 49 architecture docs, API reference (78 commands), RAG search, and a VSIX plugin. Onboarding time reduced dramatically.

C# .NET 4.0 RAG VSIX

Yumastudio — a song studio on your own hardware

Lyrics in, a sung song out — entirely on a local GPU, no cloud. Three singing engines, with quality checked automatically: the vocal is transcribed back and compared against the source lyrics.

ACE-Step HeartMuLa DiffSinger Whisper Docker

YTG — a server-side Telegram client

Building a corpus of your own voice to work with AI, plus analytics across groups and channels — from short recaps to participant profiles. Meaning-based search, with models running locally so the conversation never leaves the machine.

Telethon PostgreSQL + pgvector Ollama Docker

LOS — an AI agent platform for legal services

A general-purpose platform where AI agents do the legal work: document analysis, drafting case positions, tracking deadlines and internal policies.

Python LLM agents Docker

FootGraph — sports betting: forecasting match outcomes

Estimates outcome probabilities and compares them with the bookmaker line — the gap is the betting candidate. Facts about teams, players and injuries are stored with the date they became known, so the model never learns from what nobody knew on match day.

Python Gradient Boosting SQLite graph RENAR

Battleship

morebattle.ru

A two-player game by the classic rules: matchmaking, private rooms, three AI levels and ratings. No sign-up, straight in the browser.

Node.js WebSocket

VPN Client in Rust

ypncore.ru

Custom engine: 8 protocols, DPI bypass, ~5 MB. 4x more compact than Go alternatives.

Rust· Tauri v2· TLS/QUIC

AI Video Production Platform

Full pipeline: script → voiceover → video → editing → auto-posting. Zero manual work.

FastAPI· Celery· YandexGPT

QR Label Generator

Batch processing up to 4,480 pages. Hours of manual work replaced by minutes of generation.

FastAPI· Celery· Redis

AI Model Fine-tuning (LoRA)

Model trained on 500+ texts + RAG search. Draft articles in brand voice generated in minutes.

LoRA· Unsloth· ChromaDB

DOMUS CINEMA

domus-cinema.ru

Promo website + AI movie poster generator for a premium design studio. Multi-provider generation, upscale to A2, print-ready output.

FastAPI· FLUX· Luma AI

Crypto Trading Bot (Bybit)

Grid-trading bot with 3 AI analysts (GPT/Claude/Grok), Telegram control, emergency stoploss, and auto-reports.

Python· Passivbot· AI Analytics

WoW Chat Translator

GitHub

Real-time World of Warcraft chat translator. Addon memory reading via ReadProcessMemory, click-through overlay, <1 sec latency.

Python· PyQt6· DeepL

A2DP-Commander

GitHub

Bluetooth manager: A2DP/HFP switching, LDAC/aptX encoder in Rust, auto-mode by application.

C# WPF· Rust· .NET 8

AI Content Generator

Text and image generation via YandexGPT/ART, moderation, auto-posting to Telegram, VK, Zen.

FastAPI· React· YandexGPT

Telegram Bot Ecosystem

5 bots: Yandex Cloud billing monitor, support ticket system, AI tarot/horoscopes, recipe finder by ingredients, trading management.

aiogram· Redis· YC SDK

Akyn Bot — Chat Historian

GitHub

Telegram bot: hierarchical chat history summarization, daily digests, context search. Multi-LLM (Claude/OpenAI/Ollama).

Python· SQLite· LLM

YNS — Network Monitor

GitHub

Windows app for real-time network process monitoring. Packet capture, domain grouping, traffic statistics.

Tauri 2· Svelte 5· Rust

Offline PWA for Road Crews

12-step report form, materials and road marking tracking, 6 roles, auto-sync when network appears.

React· FastAPI· IndexedDB

AI Audit of B2B CRM

Legacy CRM infrastructure research for service industry: 9 modules (widget, SMS, WhatsApp, product marking, medical integrations) using AI tools.

PHP· Node.js· BullMQ

Turnkey infrastructure

Full cycle with AI-native engineering: design, deployment, operation — at any scale. You get infrastructure described in code and documented, not handed over verbally.

Ansible· Docker· IaC

Redmine KPI Plugins

Automated employee performance reports and intangible asset tracking. Integration into existing Redmine.

Ruby· Redmine· Rails

OSINT Geolocation Tracker

Docker-based location tracking system with 9 collection scenarios and Telegram bot management.

Docker· Node.js· SQLite

KAMRAD — offline knowledge node

GitHub

An autonomous knowledge and education server: AI chat, libraries, maps — all without internet. A derivative of Project N.O.M.A.D. (Apache 2.0).

Node.js· Docker

PDF signing service

Document signing with integrity verification, deployed as a container.

Python· Docker

Cadastral engineer toolkit

Removes the manual routine around the land registry: addresses normalised to an accepted form, cadastral numbers enriched, territory plans processed in batches. A desktop app built for the clerk, not the programmer.

Python· PyQt5

Personal money assistant

Tracks cash flow: what you have, how long it lasts, and where the plan diverged from reality. Proper accounts and transfers, with AI sorting expenses into categories.

AI categorisation· Bank sync

How I work

1

Understand the task

A call or chat. I dive into the business context, capture requirements and constraints. Decomposition, timeline, cost — fully transparent.

2

Deliver fast

I write code, set up infrastructure, deploy. Using modern tools and automation — I work as three without sacrificing quality.

3

Hand off with docs

Weekly demos. Code in your repository. Documentation, tests, and deployment included. You don't depend on me after handoff.

Career

18+ years: from Perl developer to Head of Division at Gazprom-Media

2024 — 2025

Gazprom-Media

Head of Division

Development and deployment of AI solutions for technology process analysis. Built an intelligent analytics system that improved decision-making transparency.

2022 — 2024

CYBER-ROM

Head of Infrastructure Development

Built the department from scratch (5 people). Designed PaaS platforms and Infrastructure as Code model. Infrastructure integration during M&A deals — merging systems of combining companies.

2017 — 2022

LitRes

Head of Infrastructure Development

SRE+DevOps team of up to 15, 100+ servers, 2 Gbit/s traffic. Refactored infrastructure and freed massive resources: 340 CPU cores, 2 TB RAM, 17 TB storage. Hybrid cloud, monolith → microservices, SLA/SLO for business. Speaker at DevOpsConf.

2014 — 2023

YumaLabs Group

Founder & CEO

Digital solutions for brands and agencies: promo, analytics, CRM, BTL. 100+ projects, team up to 15. Clients: McDonald's, PepsiCo, Volkswagen, Disney, EA Games via Leo Burnett, Seven, and other agencies.

2007 — 2014

Earlier

El Pako (interactive installations), KOKOC Group / Elitoid (CTO, high-load), Buongiorno Digital (Webenabling TM; P&G, L'Oreal, MTS), RBS Corporation (Senior Perl Developer).

Training and certificates

Certificate: IT Service Management Fundamentals with ITIL 4.0, Specialist Training Centre at Bauman Moscow State Technical University, 2021

IT Service Management Fundamentals with ITIL 4.0

Specialist Training Centre at Bauman Moscow State Technical University, April 2021.

ITIL 4 service management: the service model, service levels, incident and change management — the vocabulary a client uses to describe operations rather than development.

Talks & Publications

Sharing experience at conferences and in professional media

tceh 2019

Diving into DevOps Culture

Setting up cross-team communication. Talk at tceh accelerator (Moscow).

DevConf 2019

Extreme Optimizations: Working Under High Load

Talk at DevConf 2019. Video on YouTube.

DevOpsConf 2019

DevOps Fundamentals — Entering a Project from Scratch

Talk at Russia's main DevOps conference. Video on YouTube, transcript on Habr.

CNews 2024

On-premise Infrastructure vs Cloud

"Cloud and Platform Services" conference. Building a PaaS platform, comparing approaches.
Habr RU 2026 · Part 1/6

A Year and a Half Without Hand-Written Code: Why Instructions to an AI Agent Are Not Engineering Discipline

Opening piece of the series: why a rulebook is not enough and what discipline the work actually needs.

Habr RU 2026 · Part 2/6

An AI Agent Is Not a Programmer: Five Observations and Three Consequences

What an agent can and cannot do, and what follows for a team putting one into production.

Habr RU 2026 · Part 3/6

If the Agent Writes the Code, What Does the Human Become?

The engineer's role shifts from writing code to framing tasks, reviewing, and designing the agent's context.

Habr RU 2026 · Part 4/6

Specification, Gates, Metrics: How SENAR Closes the Entry and Exit of a Task

The SENAR methodology: a formalised entry, control gates, and measurable definition of done.

Habr RU 2026 · Part 5/6

The Agent's Environment: Context, Architectural Boundaries, Project Memory

Managing context, drawing architectural boundaries, and keeping project memory across sessions.

Habr RU 2026 · Part 6/6

The Limits of 100% Agent-Driven Development: Closing the Series

Where fully agent-driven development ends and what stays with the human.

Habr RU 2026

Biting into World of Warcraft Memory: A 20-Gigabyte Drift

Reading game chat out of another process's memory: region scanning ate half a core; a numeric anchor in a Lua table cut the overhead to 0.1%.

Articles marked RU are published in Russian — the six-part series on working with AI agents is complete, and all parts are linked above. Also: "Na Stachku" (Innopolis, 2020), private Rostelecom conference (Voronezh, 2019), and other industry talks.

Technology Stack

AI / ML

OpenAI API Claude API YandexGPT Ollama RAG LoRA / PEFT Unsloth pgvector ChromaDB SpeechKit

Frontend

React 19 Next.js 15 Svelte 5 SvelteKit Tailwind CSS Tauri v2

Backend

FastAPI Celery RabbitMQ aiogram .NET

Languages

Python Rust TypeScript JavaScript C# SQL Perl

Databases

PostgreSQL Redis SQLite MongoDB

Infrastructure

Docker Kubernetes Ansible GitLab CI/CD Prometheus Grafana Nginx

System / Network

Rust (async, tokio) TLS / QUIC / gRPC DNS (DoH/DoT/DoQ) WinDivert wintun / utun

Common questions

One per role — what exactly I do

When should you hire me as a developer, and when as an engineering manager?

As a developer — when you need a working product and have no team for it: I take the whole thing, from architecture to deployment. As a manager — when the team exists but the process does not hold: building a department, owning infrastructure development, setting up work with AI agents so the result can actually be verified.

How is development with AI agents different from ordinary development?

The bottleneck moves from writing code to verifying it. An agent produces plausible output faster than a human can read it, so you need formal gates: a task before code, evidence before closing, dead ends written down. I described this in the SENAR and RENAR standards and implemented it in TAUSIK, a tool that refuses to close a task without evidence.

What does AI mentoring give a team, and how long does it take?

We work on your code and your process: where an agent genuinely speeds things up, where it lies confidently, and how to tell the two apart. The outcome is a working set of rules, not a retelling of blog posts. Format and scope depend on team size and are agreed separately.