AI-native analytics, on-chain data, Telegram and agent workflows. This page is for fresh writing and practical notes from the tools I actually test. Background, work history and links live on About.
Latest Writing
- Claude Code /compact: Context Compression, Compaction Prompts, What Survives
- My Claude Code Setup: MCP Servers, Hooks, Skills and Agents (2026)
- Web Scraping AI Agents: What Actually Works in 2026
- My AI Dev Tools in 2026: What I Actually Use Daily
- Best Claude Code Skills and MCP Servers for Agent Workflows
- Show Me Your AI Setup #1: Ghostty, ownyourchat, Descript, Spokenly
What to Read Here
The homepage is a map of recent writing, not another copy of my resume. I keep it focused on tools and workflows that are being tested in actual work: AI agents, Codex and Claude Code operating loops, on-chain analytics, Telegram automation and personal systems for project memory.
Blog posts are pieces that grew out of my Telegram writing: the original idea stays recognizable, then gets enriched with context, links, examples and notes from other sources. Articles are separate search-driven topics that need a denser guide, comparison, tables, prompts and takeaways. Both formats matter: shorter posts provide context, longer articles help unpack the subject in more depth.
Some notes are short and useful as a quick reference before solving a similar task. Other pieces are longer: they unpack context, constraints, alternatives and the practical result. The simple idea is that the site should work as a useful archive of tested findings, not a display case, so a link is worth revisiting a week later or sending to someone with the same question. Everything is written for practice, not reporting.
Main Topics
If this is your first visit, start with the pieces about my AI setup, agent workflows and tool reviews that already went through real work. I try to write down concrete takeaways rather than abstract impressions: what made the work faster, where the process had to change, which settings are worth copying and which lessons should be kept for the next project.
From there, it is easiest to follow the topic pages: AI agents for practical work, Codex and Claude Code for development, TON data for research, Telegram automation for products and channels, second brain systems for personal memory. A useful article here should help with a next step: test a hypothesis, build a prototype, tune a workflow or avoid a mistake that already showed up in practice.
- AI agents — practical workflows where an agent reads context, works with files, checks its own output and carries a task through to a concrete result.
- Claude Code and Codex — skills, MCP, hooks, browser smoke checks, diff review and long tasks that need a controlled operating loop.
- TON data — on-chain analytics, Dune, research queries, product metrics and ways to turn raw blockchain data into a decision a team can use.
- Telegram automation — mini apps, bots, content pipelines, channels and operational interfaces where Telegram becomes part of the work system.
- Second brain — Obsidian, GBrain, project memory, raw notes and rules that keep decisions available after long agent sessions.
How to Use the Site
If you arrive from search on a specific article, the best next step is usually to stay with that page language: Russian material lives in Blog and Articles, English material lives in Blog and Articles. The RU/EN switcher keeps those addresses stable instead of hiding them behind a forced redirect, which is better for readers and search crawlers.
I try to publish notes after real use rather than writing abstract tool roundups: what sped up the task, where the tool broke, which settings are worth copying and which conclusions should be saved in project memory. That is why the homepage highlights the newest pieces with images, while the full archive stays in the section pages.