Allow this site to count your visits with Yandex Metrica? Metrica will store cookies in your browser to do this. How the site uses cookies
I bring AI into established businesses: online stores, clinics and service companies with 10–200 staff. I build assistants that answer customers in chat and voice bots that take phone calls. I also automate paperwork and marketing. I use similar systems every day in my own projects.
This covers everything I do, from a customer's first message to your company's own app. For every task I estimate in advance when it will pay for itself.
I built all of these for my own work and tested them there. I put together client solutions from the same pieces.
The platform launches ad campaigns in batches, checks every field before launch and sends conversions to the tracker. Every morning a summary of the previous day arrives in Telegram. It runs daily on live campaigns.
I work with any VPS and can deploy anything on it. A service that crashes restarts on its own, and logs are collected in one place. Site certificates renew automatically, and the servers are linked by an encrypted tunnel. When something goes wrong, watchdog scripts send an alert to Discord and Telegram.
Several of my systems make decisions in real time. They receive data the moment an event happens, act on it, record everything in a database and send a summary to Discord. These are fintech projects, and I don't disclose the details.
Every notable action across all my projects is recorded in a shared database. Next to it sits a knowledge graph that gives the AI assistant long-term memory. Scheduled jobs check that the records are intact, and important decisions are copied to a separate Discord channel. That is why the assistant still has the context months later.
Starting from a website address, I work out step by step how old the domain is, what the site is built on, which services are connected to it and what has changed over time. I use only legal public sources: registration records, certificates, page archives. Ready-made scripts collect the data.
Two browser profiles run in parallel and keep working after updates: a background service restores the settings on its own. The rollback instructions sit next to the code.
I wrote all the code in these systems myself. They run on my own servers and on my work computer, and I am the one who monitors them.
We talk over Zoom, Telegram or Google Meet, whichever suits you, and go through how things work today. Then I list 3–5 tasks worth automating and estimate when each will pay for itself. If I don't see a payback, I'll say so, and we can stop there.
If we decide to go ahead, I write a detailed spec: what we build, with which tools, and how it connects to your CRM and accounting. The quote is fixed, so the amount in the contract doesn't change during the work.
Once a week I send a short screen recording: what's done, what's next and where we hit difficulties. You get access to the working version straight away and can try it as the work goes on.
I stay in touch, keep an eye on the system and extend it whenever you need something new.
These are projects I can show openly. I can tell you about work under NDA on a call, without naming the clients.
Every week the accountant moved Treasury XML statements into 1C by hand. One statement took 2–3 hours.
I wrote a converter that reads both versions of the Treasury format (V3 and V4). It produces a file that 1C imports like an ordinary bank statement.
Every week the team spent 4–6 hours pulling reports for 8 clients together from Facebook Ads, Yandex Direct and Yandex Metrica.
A workflow in n8n (an automation builder) collects the numbers from the ad accounts and analytics tools. AI writes notes on unusual changes. The finished report goes to the client as a PDF or a Notion page by 9:00 on Monday.
The owner's working sheet, the warehouse book and the CRM each showed a different revenue figure. Site revenue in the owner's sheet differed from the CRM by 8.4%, and no one knew where the gap came from.
First I compared the three sources over the same period and traced every gap to specific orders. Almost all of the 8.4% came from cancelled and partially refunded orders. Then I turned the check into a daily job with a 3% threshold.
My name is Vyacheslav, and I have worked in IT since 2017. I started as an engineer at the Baltic Shipyard, building multi-level production schedules in Primavera, a project planning tool. Then I moved on to Infor LN, an ERP system that runs manufacturing, and customised it in BAAN 4gl: first at the shipyard itself, then at IPL Consulting in Turkey and Asertiva Solutions in Bulgaria. Clients included carmakers, companies in the nuclear and shipbuilding industries, and plants in Russia, the EU and the Middle East.
Alongside that I studied at App Brewery: first the iOS & Swift Bootcamp, then web development. I built 20+ iOS apps as study and personal projects (Swift, SwiftUI, CoreML, ARKit) and took part in WWDC21, Apple’s developer conference.
In 2024 I took up media buying: first as a mid-level buyer at a large media-buying team, then as a senior buyer at a second team. Both names are under NDA. Over that time I managed $1.34M in ad spend, with $379k in net profit. That is also when I started writing my own tools: automatic campaign launch through the Facebook Graph API, a collector for competitor ads, and sorting of incoming requests with GPT.
Today several of my own AI services run around the clock, covering fintech, client acquisition and routine automation. I bring the same approaches to small and mid-sized businesses. I take on the whole project: I write the code, launch it and support it so it still works a year later.
These are the formats I work in, with starting prices. The final amount depends on the scope. We set it in a quote once the written spec is ready, and I write the spec for free. Prices for individual services are in the “What I do” cards above and in the table on the Pricing page.
A good place to start. On a call we go through how your work is organised. Afterwards you get a list of tasks worth automating and a payback estimate.
One process, end to end. We test the approach on a specific task before investing in a bigger system.
A complete system: several channels, a connection to your CRM, and training for your team.
After launch I keep everything running and extend the system as new tasks come up.
I work as a self-employed contractor under a civil-law contract.
What a booking chatbot is made of, where it breaks, and what customer data the law lets you send to an AI model.
What an AI receptionist and booking chatbot is made of, the four places it breaks in real use, and nine questions to ask a vendor before you pay.
Read the articleCan you send customer data to ChatGPT? What 152-FZ requires, why anonymisation often fails, and how to check your own system in 20 minutes.
Read the articleI'll tell you what to automate first, when it will pay for itself and whether your task needs AI at all. Sometimes it doesn't.
Or just ask a question. During business hours I reply within 2 hours