By day I'm a knowledge manager: I train an AI support agent and write documentation that both people and LLMs understand. When something slows the work down, I build a tool with Claude. Here I write about how it turned out.
What I create and maintain: from SaaS to small tools for my own work.
Telegram bot for paid access to private channels and groups. A flat plan, 0% TribeSync commission on sales, payments in crypto and Telegram Stars.
Content engine for solo founders and SMM: brand profiles, AI post ideas, weekly content plans and posts written with proven copywriting formulas.
Website of my translation bureau, moved from WordPress to Next.js.
Chrome extension for annotating screenshots right on the live page: boxes, tooltips, arrows, redaction and export.
Tool for tour guides: automated notifications, flight logistics and group management.
Internal dashboard for a help center: syncs live articles and surfaces content-quality and translation gaps.
Recent notes, ideas, and discoveries.
How I turned Ultimate Content Machine from a browser-only generator into a content store Claude can read and write through MCP, with a fact-check gate before anything gets approved.
Why help center articles now have two readers, a human and an AI agent, and how to structure support content so the machine can retrieve the right answer.
How to leverage Google's NotebookLM as a documentation auditor to stress-test your SaaS help center, find edge cases, and empower your support team.
The work that sparks ideas for posts.
I run the product's knowledge base and train the AI agent that answers customers from it: I read real conversations, find what went wrong and fix the cause at the source.
I design documentation so that both readers and LLMs find the right answer. I also review and write UI copy for product interfaces.
Glossary, translation memory and a CAT workflow, then an AI agent to scale and automate the process.
When a task eats too much time, I build a tool for it with Claude: dashboards, browser extensions, Telegram bots.