I build the systems that answer every lead, reply and quote: quote follow-up, outbound that reaches the inbox, instant lead response and booking reminders. Every system I ship tells you the moment something breaks.
My clients don't share an industry. They share a problem: work that depends on someone remembering, copying or checking by hand. These are the situations I'm called in for most.
Calls after hours, emergencies that can't wait until morning, and a front desk that can't answer everything. I build phone agents that triage, book into the practice system and reach the right person.
Seen in: veterinary phone intake
Pipelines, follow-ups and CRM syncs that don't drop or double people, and newsletters that land in the inbox instead of junk.
Seen in: nurture engine, CRM sync, deliverability
Quotes that go out and never get chased, a shared inbox, customers asking where their order is. I run this work myself at a small B2B equipment importer.
Finding the businesses, finding the person, sending, and handling the replies, run as one system and watched around the clock.
Seen in: outbound engine
The spreadsheet that runs everything, turned into a proper database with forms, a dashboard and a weekly report that sends itself.
Seen in: grant pipeline, venue operations
Multi-tenant apps, billing, deploys and the automated tests that keep them standing once real customers arrive.
Seen in: veterinary platform, public code
Most of my client work is private, so clients are described rather than named unless they've been public about working with me. Every detail below is from the real build.
A phone agent that answers a clinic's calls, works out how urgent the problem is, and books the visit or gets a vet involved. I built it and rebuilt it as the company's own platform.
The firm wins its clients by email. I built the whole chain, from finding the businesses to answering the replies, so the team's time goes on booked calls.
Replies don't wait for office hours, and a wrong automatic answer to a prospect can't be taken back. So the worker answers only the clear cases itself, and the sorter is held to a fixed test set.
Python · Cloudflare Workers · Instantly · Apollo · Hunter · Supabase · Google Sheets · Slack
When a deal goes into nurturing, the advisor should hear about it again at the right time, and never about a deal that has already moved on.
The re-check after the wait. Without it, an advisor gets a task about a deal that closed three months ago, and the team learns to ignore the task list.
GoHighLevel workflows · pipeline stages · tags · tasks
Contacts lived in GoHighLevel and in Kit, and someone was copying them between the two by hand. I replaced that with a sync in both directions.
Two systems that are both allowed to write the same record will keep copying each other's changes. The sync has to tell its own writes apart from real ones.
n8n · GoHighLevel API · Kit API · Python
In both cases the authentication records were already clean. The fixes were in which systems were sending, and who the mail was going to.
Google Workspace · GoHighLevel · DNS · SPF, DKIM, DMARC · Postmaster Tools
Authentication-Results: spf=pass dkim=pass dmarc=pass Records clean. Look elsewhere: sending systems, sending identity, and who is on the list.
Applications arrived through a form and lived in a spreadsheet. Six days from kickoff to walkthrough, the board had a database instead.
The same charity turns up under slightly different names from one year to the next. Matching on tax ID first, then website, then name keeps one charity as one record.
Airtable interfaces and automations · form webhook · Python import scripts
A church building with rooms hired out through the week. For almost three years I designed and ran the systems behind it.
HTML, CSS and JS · GitHub Pages · Python
Room pages, blog and safeguarding, rebuilt from scratch and handed over.
louiskowal.github.io/lurc-connectIt interviews you about your business, then switches into selling your offer to you. A browser preview of the agent logic, built to run on WhatsApp Business.
Message itA multi-step eligibility check scored by a rules engine against five finance products. The AI only explains the result. It never makes the decision.
Take the assessmentTrades, dental, real estate, legal and hospitality, each with its own way of sorting urgent from routine. Tell me your business and I'll send the number for yours.
Ask on a callTenant isolation in Postgres and the four ways it silently fails, tested against a real Postgres 16 in Docker. Audit logging in a trigger, and migrations that don't take the app down.
github.com/louiskowal/multitenant-postgresTwo-way sync without the duplicate loop: deterministic identity keys, echo suppression and a dead-letter table. No dependencies.
github.com/louiskowal/sync-ledgerReference implementations, rebuilt clean. No client code and no client data.
What sets Louis apart is his ability to understand complex systems quickly and translate ideas into clean, functional implementations without unnecessary friction. … I never felt like I had to micromanage him. Louis takes ownership and executes.
“Louis communicated great, very nice and kind, and he went above and beyond when I asked him. He didn't have to but he did. Would def hire again.”
“What impressed me most was his transparency.”
I'd rather tell you not to automate something than build the wrong thing well.
One call about the part of the business that costs the most. If automation isn't the fix, or the task takes an hour a week, I'll say so.
Before anything runs on its own, I run it once on your real data and we check the output together.
Every step that can fail gets its own alert: what broke, why, and what to do, sent where you already look.
It runs in your accounts, with written documentation. You own it, and you can call me when you want more.
Each of these comes from a system that broke, nearly broke, or would have broken quietly.
Wrong output running for three weeks costs more than a clean stop. So every step that can fail sends an alert.
Leads arrive from several places, so the same person enters twice. Two reps calling one lead is worse than a slow follow-up.
Prices, capacities and emergencies are decided by plain code. The model handles the conversation.
A reminder about a deal that closed three months ago teaches the team to ignore the system.
Some spreadsheet appends count the rows, then write. Leads that arrive together collide and one disappears without an error. I test with simultaneous submissions.
Gmail places mail reader by reader, from each reader's history with you. Who you send to matters as much as your records.
It collects, triages and books. Prices and promised times come from a person, which keeps a real deployment out of trouble.
Money, legal documents, messages to customers and ad spend wait for an approval step.
I'm French, I live in the UK, and I work in English and French. I started LKAUTOMATE in 2023.
Most of what I build has to keep working after handover, with nobody watching it. That's why every system I ship comes with alerts and written documentation, and why I test against real data before anything goes live.
Alongside client work, I spent almost three years running the operational systems for a community venue in Loughborough. I've been classically trained on harpsichord and organ since I was five, and I'm an Ironman finisher. The discipline transfers more than people expect.
I work alone or with a small circle of people I trust. Clear scope, fast iteration, no surprises, no upselling. I'd rather be the second AI builder you hire, and the last.
A 30-minute call, no prep needed. If there's a fit, you'll get a clear scope and a price. If there isn't, you'll leave with at least one useful suggestion.
Book a 30-minute calllouis@lkautomate.com