Sergey Vareyko / AI Automation

AI automation for business: AI receptionists, chatbots and content pipelines

This page is for businesses that want an AI receptionist or chatbot that answers clients and books appointments, for companies that already produce material worth publishing — videos and photos of their work — but have no one to turn it into a steady blog, and for teams that want AI inside their own product rather than in yet another subscription. I'm Sergey Vareyko, a software architect with 15+ years in Java and Spring. The AI automation for business I build is practical: models answer, analyze, write and check, a person decides or takes over when a question is not standard, and the system does the rest on a schedule.

What I build

Most of what follows comes from Blog Writer, a self-hosted AI content pipeline I designed as its solution architect.

  • An AI receptionist that books. A chatbot in the web chat and in Instagram Direct that answers clients' questions, books appointments through the platform's own scheduling engine and hands over to a person when the question is not standard, as in Smart Reception below.
  • Pipelines that start from your own material. The system watches the company's Telegram and Instagram, fetches every new post whole — video, photos, caption — and accepts manual uploads.
  • Video and image analysis. A model watches the whole clip and extracts a structured record: for that client, the car, the service, the before-and-after measurements and the moments worth a frame. Frames are scored, a hero image is chosen, and any picture can be redrawn from a description in any language.
  • A writer, an independent critic and a human. The writer drafts a full SEO article to a contract: title, meta, structure, FAQ, internal links. A programmatic SEO check and a critic model then verify facts against the source, technical consistency, overlap with what is already published, and whether every operator comment was honored. After one automatic revision a person approves, edits, comments or rejects; nothing goes live without approval unless AUTO mode is on.
  • Publishing into the CMS you have. Bitrix through a one-file PHP bridge, with publication slots, a weekly cap, sitemap refresh and cross-links from older articles to new ones.
  • Research from real search demand. Topic proposals from Google Search Console, Yandex Webmaster, Yandex Wordstat and search suggestions, and from clusters of finished jobs; new articles cite only the company's relevant cases.
  • Any model in any role. Anthropic Claude, OpenAI, Google Gemini or any OpenAI-compatible endpoint, including a self-hosted model, assigned per role (writing, review, research, vision, video, transcription, images) and per site, and switched in settings without code changes.
  • Cost and load under control. A paced queue with an hourly budget and a buffer of ready articles lets a backlog of hundreds of old posts drain on schedule instead of flooding the reviewer; cost is tracked per article.

Beyond content pipelines, I add AI features to products such as Bolimond and Tobo, covered under content websites and e-commerce.

How I work

An AI pipeline is still software: it needs requirements, an architecture and a person accountable for what it publishes. I start from what you already produce and where it should end up — which channels, which CMS, who approves, in which languages — and define the structure, the model for each role, the integrations and the limits on cost. Then a working version runs on your real material, and we iterate on the contract, the critic's checks and the schedule until the output is something you would sign.

I use AI tools — Claude, Cursor, ChatGPT, Ollama — in my own engineering work, and the systems I build keep a person in the decision. You receive a self-hosted system with an admin panel (Blog Writer's is bilingual, English and Russian), encrypted secrets, and nightly encrypted off-site backups with one-click restore — the same operations side as my internal tools. The same building blocks apply when AI has to be added to an existing Spring application.

Projects in this domain

Smart Reception — Architect and Engineer, 2026 to date. An online booking platform whose AI receptionist answers clients in the web chat and in Instagram Direct, books appointments through the platform's own scheduling engine, and hands over to a person when the question is not standard; an AI assistant in the admin panel writes service descriptions, turns a photographed price list into services and summarizes reviews for the owner. A self-hosted model behind a gateway with a cloud fallback, so a salon's data stays on the platform's own servers.

Smart Automation — Author, 2026 to date. The site of an AI automation practice for small and medium businesses: twenty-plus prerendered pages in Russian and English on what can be automated (lead handling around the clock, document flows, chatbots, CRM integration, analytics, sales forecasting) and for which industries, a lead form with instant Telegram and e-mail alerts, an admin panel with visitor analytics and scheduled encrypted backups, and SEO titles and descriptions drafted by a model from each page's own text. One Spring Boot jar with the Nuxt front-end built into it.

Blog Writer — Solution Architect, 2026. An AI content pipeline that keeps a company's blog alive from the videos and photos it already posts about its work, built on Java 21, Spring Boot, MongoDB, Thymeleaf and htmx, with ffmpeg and ImageMagick for media.

Bolimond — Author, 2026. A portfolio and CMS for the artist Constantin Bolimond with a GrapesJS visual page builder; it includes internal AI features built on an in-house AI client.

Tobo — Author, 2026. A single-JAR storefront and admin CMS for a handmade ceramics brand, with AI integration to simplify the shop's processes alongside dynamic SEO and analytics.

TouchCommerce — Software Engineer, 2010–2016. Not AI, but relevant if you are thinking about a chat assistant: a customer engagement platform for large brands with live chat, guided self-service and automated engagement. I built backend and frontend for its real-time chat, including co-browsing, and automated most of the ticket resolution workflow.

Frequently asked questions about AI automation for business

Can you build an AI chatbot for my business?

Yes. The AI receptionist in Smart Reception is a chatbot of exactly this kind: it answers clients in a web chat and in Instagram Direct, books appointments through the platform's own scheduling engine and hands the conversation to a person when the question is not standard. It runs on a self-hosted model behind a gateway with a cloud fallback, and I built real-time chat earlier at TouchCommerce. Whether a chatbot is the right tool for your business depends on what it should answer, from which data, and who is accountable when it is wrong.

Does my data have to go to OpenAI or another cloud provider?

No. Every role can point to a cloud provider or to a self-hosted model behind an OpenAI-compatible endpoint, so you can keep analysis on your own hardware and use a cloud model only for writing. The pipeline itself runs self-hosted under nginx and systemd.

How much does it cost to run?

It depends on the models you choose and the volume. The system tracks cost per article and works through an hourly budget, so spending stays where you set it.

How long does setup take?

It depends on scope: how many sources and sites, which CMS, how many languages, and how strict the approval flow is. A new CMS means a new bridge, like the one-file PHP bridge written for Bitrix.

Need an engineer with experience in AI automation for business? Write to me — a few lines about the task are enough — or look through my CV.