oleksandr@prod:~$ whoami

Oleksandr Makhinko — AI Automation Engineer · Ukraine · remote, worldwide

AI agents and automations that remove manual work from real business processes.

I help e-commerce and service businesses stop losing leads and drowning in support — with AI agents that answer in seconds, 24/7.

n8n · AI agents · RAG · CRM & API integrations

Built production-grade: error handling, monitoring, security — from day one.

Discuss your automation See the work

FIG. A — typical build: lead intakedrawn as built

systems built to production standard

15+

integrated services

try it live

submit a test lead — watch it hit 4 systems ↗

work — featured systems

Selected systems, drawn as built.

Demo builds around realistic business scenarios — each one taken all the way to production standards, not to a screenshot. Every diagram matches the running workflow; metrics are measured in test or modeled from the scenario.

live demodemo build

AI Sales Ecosystem — chatbot qualifier + sales analyticschatbot qualifier + sales analytics

Coffee e-commerce. An "AI barista" on the storefront qualifies visitors with BANT, quotes live WooCommerce prices, and answers from a pgvector knowledge base — with daily AI dialog analysis, a weekly CEO report in Telegram, and a voice channel on the same backend.

FIG. 01 — sales ecosystemshop.mahinko.com

answers visitors in ~5–10 s, 24/7
leads scored 1–10 straight into Zoho CRM
closed feedback loop: prompt improved from its own analytics — +75% lead score on objections

n8nClaude via OpenRouterpgvector / NeonWooCommerce RESTZoho CRMTelegram

Read case study shop.mahinko.com

demo build

AI Call Processor & Analytics Agentevery call transcribed, classified, queryable

Cake manufacturer, ~200 support calls a day. Calls are transcribed with PII redaction, classified by an LLM into a structured data table — and a chat agent on Telegram and web answers "how many negative calls this week?" in seconds.

FIG. 02 — call analytics pipeline100% coverage

call → classified in ~30 s (was ~7 h)
100% call coverage (was 30–40%)
self-healing: sweeper + error alerts

n8nAssemblyAIGemini · OpenRouterLangChain agentGoogle Sheets

Read case study

demo build

AI Ticket Routing + SLA Controlinbox triage with a chain of escalation

E-commerce support inbox, ~350 emails a month. A chain-of-thought classifier assigns category, P1–P4 priority, owner and SLA deadline; a 15-minute watcher escalates L1 → L2 → L3 all the way to the CEO.

FIG. 03 — routing + escalationzero lost VIPs

email → routed in ~7 s
12/12 correct on the test set
3-level escalation — zero lost VIP requests

n8nGPT-4o-miniGmail APIGoogle SheetsTelegram

Read case study

live demodemo build

AI Lead Qualification & Multi-Channel Funnelone form in, four systems updated

Beauty salon. One landing form fans out through a webhook to Google Sheets, Telegram and Zoho CRM; AI scores intent hot / warm / cold and drops each lead into the right Klaviyo list for personalized email.

FIG. 04 — multi-channel funnel4 systems, 1 form

lead processed in ~8 s (was 10–15 min)
4 systems + AI from one form
every branch isolated — one failure never blocks the rest

n8nOpenRouterKlaviyoZoho CRMTelegram

Read case study everbeauty.mahinko.com

All case studies — soon

the index grows every couple of weeks — new builds are added as they ship

services — what I build

Four things, done properly.

RAG · tool calling · memory

AI chatbots & agents

A first line that answers, qualifies and hands off around the clock — grounded in your data, not in whatever the model feels like saying.

for: store owners · coaches & experts

transcription · classification · dashboards

Call & email processing pipelines

Every call and message transcribed, classified and in a table before a human would have pressed play.

for: local businesses · support-heavy teams

webhooks · CRM · email marketing

Lead funnels & CRM integration

One form submission updates every system you run — CRM, sheets, email lists — and pings your phone on the hot ones.

for: B2B sales teams · service businesses

chat agents · scheduled reports

Internal analytics agents

Ask your business data questions in plain language — and get the weekly numbers without having to ask at all.

for: owners who want numbers, not busywork

approach — engineering standards

What "production-grade" means here.

Every system above ships against the same spec sheet. It is the difference between a demo that works once and a system a business runs on.

$ run preflight --env productionmakhinko/standards · ES-01…06

[ ok ]ES-01Error workflowsEvery failure is captured, alerted to Telegram, and retried where retrying is safe.

[ ok ]ES-02Idempotency & atomic locksRetries and race conditions can't duplicate a record or double-send a message.

[ ok ]ES-03Webhook security & PIIAuth on every endpoint; personal data redacted before it leaves the perimeter.

[ ok ]ES-04Self-healing health checksScheduled sweepers pick up anything real-time processing missed.

[ ok ]ES-05LLM cost guardsSpend caps and anti-spam limits sit in front of every model call.

[ ok ]ES-06Division of laborCode counts the numbers; the LLM interprets them. Never the other way around.

6 checks passed · 0 warnings · ready to ship

process — how we'll work

From a described problem to a running system.

No discovery-call theater. The whole path is five steps, and you see the plan and the price after step two.

OP-01Describe the process

A few sentences in Telegram or the form below: what eats the hours, where the leads go missing. Plain words — no spec needed.

OP-02Map & estimate

I take the process apart and come back with an automation plan: what to build, which systems it touches, what it saves — and a fixed quote.

OP-03Build

Production-grade from day one: error paths, monitoring, security, cost guards — the same ES-01…06 standards you see above.

OP-04Test on your real data

The system runs on your actual calls, emails and leads while we watch it together. Metrics before go-live, not promises.

OP-05Handover & support

Documentation, monitoring and a guided handover — your team can run it without me. If something breaks later, the system pages me before you notice.

// every system above went through exactly this pipeline

about

Engineer first.

I build automation the way software is supposed to be built: starting from the problem — where the hours actually go — and finishing with a production-ready system, documented and handed off. Not a prototype that works while I'm watching it.

The systems above are demo builds around realistic scenarios, and you can inspect or try them. The standards they're built to are the ones I ship to clients with: error handling, monitoring, security, cost control — from day one, not as a phase two.

location: Ukraine (UTC+2) · remote, worldwide
stack: n8n · LLM APIs · RAG · CRM & REST integrations
github: workflow JSONs + READMEs — profile in preparation
status: open to projects and full-time roles

certifications

CRT-01AI Automator (243 hours)GoIT · ID 47572 · 2026verify ↗
CRT-02Google AI Professional CertificateGoogle · Coursera · 2026

portrait · 1:1 · photo to follow

faq — before you write

Fair questions, straight answers.

QA-01How much does it cost?

Fixed quote after the mapping step — you know the full price before I build a single node. Running costs (LLM calls, transcription, APIs) stay on your own accounts, visible line by line; for the systems above they land in the tens of dollars per month, not hundreds.

QA-02How long until it works?

It depends on scope — after the mapping step you get a concrete timeline for your system along with the fixed quote, before any building starts. And you always see the system running on your real data before go-live.

QA-03What access do I need to give you?

Only the systems the automation touches — scoped API keys and credentials you create, never your personal passwords. Everything is revocable by you in one click, and secrets live in n8n's credential store, not in the workflows.

QA-04What happens when something breaks?

Every system ships with error workflows and self-healing checks: failures alert me in Telegram with full context, stuck jobs are re-driven automatically. That's not an add-on — it's the ES-01…06 standard above.

QA-05Who owns the workflows?

You do. Everything runs in your n8n instance (or one I set up on your hosting), under your accounts and your credentials. You get full exports and documentation — no vendor lock-in, including on me.

QA-06Can my team maintain it without you?

Yes — that's the point of the handover: documentation and a walkthrough of how the system works. n8n is visual, so your team sees every step of the logic instead of reading someone's code.

QA-07Will it work with my tools?

If it has an API — yes. The featured systems already integrate 15+ services, from CRMs and e-commerce to transcription and email marketing. Odd legacy tools usually still have email or webhooks, and that's enough.

contact

Have a process that eats your team's hours?

Describe it in a couple of sentences — I'll reply with an automation plan: what to build, what it touches, and what it will save.

telegram@shuriken_86 mailtomahinko86@gmail.com

responseusually within one working day