- Context
- Decision
- Handoff
AI agents
I design agents to qualify leads, answer customer questions and search internal knowledge. I define when a person needs to take over.
Explore this serviceAI SYSTEMS / AUTOMATION / CRM
MORE FREEDOMTHROUGHBETTER SYSTEMS
I design and build AI systems, CRM automations and connected workflows that give your business more room to grow.
THE SYSTEMS I WORK WITH
- n8n
- GoHighLevel
- HubSpot
- Make
- Supabase
- OpenAI / GPT
- Anthropic / Claude
WHAT I DO
YOUR TOOLS, WORKING AS ONE SYSTEM.
I start with the process you want to change.
Explore services- Lead
- Pipeline
- Follow-up
CRM & revenue systems
I configure CRM pipelines, lead routing, follow-up and reporting around your sales process.
Explore this service- Source
- Queue
- Destination
Automation & API integrations
I connect systems through APIs and automate data transfers. I define how failed requests and retries should be handled.
Explore this service- Boundaries
- Data flow
- Responsibility
Architecture & consulting
I review the current process and design the data flow, system boundaries and implementation plan before development.
Explore this serviceExample architecture
From trigger to trace.
An AI agent classifies a new inquiry and picks the workflow. Deterministic steps write it to the CRM, verify the record and log every step; a failed check goes to a person. This is an illustrative architecture, not a live client system.
You're the client. Write to the business
Demo: simple rules in your browser detect the intent. The text is never sent or stored.
The system handles it — no person needed
Here you'll see what the system does at each step.
- AI decides the route
- Workflows execute
The result for the owner
Pick an example or write your own message above — what got done without you appears here.
Trigger
What happens
A form submission, email or webhook starts the workflow. The input is captured exactly as received.
If it fails
Empty, malformed or repeated events are rejected before anything is written.
AI agent
What happens
A language model reads the inquiry, classifies it and picks the workflow to run. It makes the routing decision; deterministic steps do the writing.
If it fails
If the model is unsure or unavailable, the inquiry takes a safe default route and is flagged. The agent never writes to the CRM directly.
CRM
What happens
The contact is matched against existing records, then created or updated without duplicates.
If it fails
If the CRM does not respond, the write is retried with growing delays. It never fails silently.
Validation
What happens
The saved record is read back and compared with what was sent.
If it fails
A mismatch goes to a person for review instead of being guessed or overwritten.
Human review
What happens
The record is held and the owner gets the case with full context to decide what to do. Nothing is overwritten automatically.
If it fails
The case stays open in the log until someone resolves it.
Database
What happens
Every step leaves a trace: what happened, when and with which outcome.
If it fails
The trace is how problems are found and fixed later, without retyping data by hand.
ABOUT
BEFORE TECH, I RAN HOSPITALITY, DELIVERY AND B2B TRADING.
Before tech, I ran hospitality businesses, food delivery operations, and sales. I've seen what a broken process costs: time, money and real customers.
- Role
- Serhii Keiko, marketing automation & CRM systems architect
- Before tech
- Hospitality, food delivery, B2B product trading
- Upwork
- Top Rated
- Teaching
- Author, speaker and curator of the AI agents course at the MASC Space online school
SELECTED WORK
SYSTEMS IN PRACTICE.
HIPAA-grade integrations for a US psychiatric practice
Spruce · Zoho Deskn8n in the client's AWSAudit log · 7 detectorsGoHighLevel · DrChrono · Zendesk
I connected the practice's patient messaging, CRM, medical records and support desk in its own AWS account, with an audit trail that cannot be edited and alerts that catch failures before staff do.
Read the caseAI receptionist and patient journey for a dental clinic
Chat and SMS · Telegram · Web chatGoHighLevel AI · n8n agentContacts · Calendar · Pipeline and tasksBooked visit · Task for staff
A clinic system where an enquiry turns into a booked appointment without staff, with an AI receptionist in GoHighLevel and a second agent in n8n that works with the CRM through four MCP servers.
Read the caseApplication intake and multi-lender CRM for a property-purchase service
Webflow formGoHighLevel contactLender A · Lender B · Lender CKit sequences
One application on the website now creates the client, a separate deal for every lender they chose and the right email sequence, with no retyping.
Read the caseSales and onboarding system for a senior tech-support service
For myself · For someone elseStripeGoHighLevelEmail series · Booked session
I built the whole sales path in GoHighLevel, from the first page to booking instructions, with a separate gift flow for people who buy help for a parent.
Read the case- 01
Audit
Where time is lost
What you getFindings and priorities
- 02
Architecture
Data flow and boundaries
What you getArchitecture outline and diagram
- 03
Build
Small, testable steps
What you getConfigured workflows
- 04
Validate
Failure cases included
What you getTest results and known limitations
- 05
Handover
Instructions and access
What you getOperating instructions and handover
LET’S TALK
WHAT DO WE AUTOMATE FIRST?
Describe the one process that takes the most time. I'll reply within 24 hours and suggest the first step.
Discuss your projectLESS MANUAL WORK. MORE LIFE.