Your team burns 30 hours a week on work I automate
Senior AI Automation Engineer. I build n8n workflows, RAG search over your own documents and multi-agent systems that run around the clock without babysitting. Over ten products run on my own servers and stayed available more than 99.5% of the last 30 days. Your team stops shuffling data and starts doing work only they can.
Describe one process. Within a day I say whether it can be automated, what it takes and what it costs. No pitch, no commitment.
10 applications in production
Ukraine · Europe · Remote
Ihor Kostyniuk
Senior AI Automation Engineer · LLM agents, MCP, RAG, Voice AI
10+
Apps in production
2.5y+
Experience
24h
Response
WHAT I DO
Services that actually move the needle
What you get, how it runs and what it costs. The projects behind it are in the next section.
Workflow Automation
Your systems exchange data on their own. Built on n8n, running on your server or mine, with monitoring wired in: a broken flow raises an alert instead of going quiet.
n8n on your server or on mine
400+ ready integrations, the rest over API
Uptime monitoring and failure alerts
from $150
AI Agents & RAG
An agent that answers from your documents, not from the internet. Search over your own knowledge base, calls to your own tools, answers checked by a judge model before they reach anyone.
Qdrant or pgvector over your knowledge base
Tool calling and custom MCP servers
Several providers with automatic fallback
from $300
Hiring & ATS
Screening stops being a manual step. Vacancies and candidates arrive on their own, CVs are parsed, fit is scored, the technical interview is generated.
Job collection across 17 boards
CV parsing and fit scoring
EU AI Act: DPIA, DSAR, human in the loop, bias monitoring
from $150
SaaS & Custom Apps
The whole product: web, Android, desktop and the server under them. Multi-tenant with data isolation, auth, billing, over-the-air updates.
Next.js + Supabase, RLS on every table
Native Android (Kotlin) and desktop (Tauri)
Docker, monitoring, nightly backups
from $500
REAL PROJECTS
Cases that speak for themselves
Every number below comes from my CV, which is open in full.
Consolidated production into 6 reusable pipeline skeletons and grew it from 3 to 12 client channels. Renders run for hours per job: checkpointed state so any stage resumes after a crash, retry with backoff, circuit breakers around each provider. Then handed rendering over to the clients — bundled installer, deploy keys, one-click update — so they run it on their own hardware.
3→12
Channels
$22-30
Per film
~235
Scenes each
Client · since 2024
AI support across 7 channels
WhatsApp, Telegram, Viber, Instagram, Facebook, web and email all reach one agent. RAG is separate per tenant. Cleared Meta App Review for production messaging permissions. The site widget speaks the host page's language, is screen-reader accessible, and writes a visitor who leaves a phone number straight into the CRM.
7
Channels
Meta
App Review
RAG
Per tenant
Client · recruiting
Recruiting CRM under the EU AI Act
Multi-tenant CRM: 33 migrations, RLS on every table, a subdomain per workspace, around 20 AI tasks including CV parsing, few-shot-calibrated ATS scoring and outreach generation, fed by a 17-source scraping pipeline. Compliance built in: DPIA and DPA templates, self-serve DSAR export and erasure, human score overrides with an audit trail, a bias-monitoring dashboard.
17
Job sources
~20
AI tasks
86+
Security tests
Client · 2025–2026
Documents and operations
Audited four business functions — procurement, accounting, legal, recruitment — turned the bottlenecks into implementation specs, then built the systems those specs described. Contracts and reports are assembled from templates with docxtemplater and Gotenberg: drafting became filling a form. Plus the operations app the teams work from: scheduling, run monitoring, work hours and spend in one view.
4
Functions audited
SmartSuite
System of record
Docker
Under all of it
Own product · in production
Monitoring that watches itself
One view over 9 servers and 36 watchdogs. A dead-man switch raises an alert when the monitor itself dies, and a native Android app delivers push alerts. The same approach is in the Managed package: when a client's flow goes quiet, I know before the client does.
9
Servers
36
Watchdogs
Android
Push client
Own product · in production
Content pipeline with one-tap approval
LinkedIn, Instagram, Telegram and Threads from one pipeline. Every post is fact-checked against 18 sources and cross-checked by two models, and goes out only after one tap in Telegram. Routing across providers with automatic fallback keeps the daily cost near zero. The same can be built for your personal brand.
4
Channels
18
Fact-check sources
2
Models cross-check
RECOMMENDATIONS
What the people who worked with me wrote
Three recommendations from LinkedIn, quoted word for word. They stay in the language they were written in.
AM
Alex Merchenko
Head of Sales & Business Consultant · Sales Strategy & Audits
Client
“We had Ihor build a lead qualification bot for our agency. Before that, our sales guys were jumping on every call, half of them going nowhere.”
“The bot now handles first contact. Asks the right questions, scores the lead, routes the serious ones. By the time a salesperson picks up, the conversation is already half done.”
“Lead qualification time dropped around 60%. That tracks with what we saw. The team stopped burning afternoons on dead-end calls and actually started closing.”
“Delivered clean, no drama after launch. Would recommend for any sales team that’s tired of chasing unqualified inbound.”
May 2026 · LinkedIn
~60%
less time qualifying a lead
The client’s own figure, from the recommendation above
OV
Oleksandr Vashchyshyn
AI Engineer
Team lead
“I was Ihor’s team lead and watched him working across several internal projects. Document generation, ATS pipelines, workflow integrations. He handled all of it.”
“What I noticed pretty quickly was that he didn’t need handholding. You’d give him a task and he’d come back with it done, plus a few things you hadn’t thought to ask for.”
“Infrastructure side was solid too. Docker, self-hosted setup, monitoring. Stuff just worked. And when something went wrong, he usually knew about it before anyone else did.”
“Good automation engineer, easy to work with. Would recommend him for any team doing serious AI automation work.”
May 2026 · LinkedIn
PS
Parsla Sietina
Business Analyst · Process improvement & analytics
Colleague
“As a colleague he demonstrated a strong analytical mindset, exceptional problem-solving skills, and a genuine willingness to help others. No matter how complex the challenge was, he approached it with patience, curiosity, and determination to fully understand the root cause before finding the right solution.”
“He has a great ability to break down complicated processes, think critically under pressure, and come up with practical solutions to issues that initially seemed impossible to resolve. He was always approachable, collaborative, and ready to support colleagues whenever needed.”
Prices in US dollars. Pay by card in hryvnia (the payment service fee is included in the amount) or by bank transfer.
Work beyond the package hours is $35 an hour, only after we agree. Subscriptions are monthly: cancel any time before the next month starts. Three months paid upfront: 10% off.
Docker Compose · Caddy · self-hosted VPS · n8n in queue mode · uptime monitoring and error tracking · nightly backups · EU AI Act and GDPR obligations
ABOUT
Ihor Kostyniuk
Senior AI Automation Engineer · LLM agents, MCP, RAG, Voice AI
I own the whole system, from the screen to the server
Started in marine engineering, where a system either works or someone gets hurt. That obsession stuck. In AI automation the bar is the same: staying up is the baseline, not a brag.
Two and a half years shipping production LLM systems on self-hosted infrastructure. I write the orchestration layer myself instead of adopting a framework: tool-calling loops, custom MCP servers, multi-provider routing with automatic fallback, checkpointed pipelines that resume after a crash, human-in-the-loop gates. End to end means the web app, the native Android app and the server under them.
English C1Ukrainian C2Russian NativeRemote-firstOdesa, Ukraine
Describe the problem, even a rough idea. I'll tell you if it's automatable and roughly what it takes. No pitch, no commitment. Usually reply within a few hours.
Message sent!
I'll reply where you left your contact, usually the same day.