AX

Selectively available for engineering work

AXISROW

Python engineer building AI agent tooling, automation, and data systems — built to survive production.

View selected work Get in touch
Stars across GitHub
136
Merged upstream PRs
59
Starred projects
16

01 Momentum

Stars over time

Cumulative GitHub stars across all repositories since 2026-03-01.

136★2026-09-24 · all repositories
Cumulative GitHub stars since 2026-03-01 The chart ends at 136 stars on 2026-09-24. 0 70 140 Mar Apr May Jun Jul Aug Sep

02 Selected Work

Systems with measurable output

Original repositories with the stars — see all at github.com/axisrow.

Integrations

tg_content_factory★ 31

Content-factory pipeline for Telegram — flagship project.

View repository ↗
hhru★ 27

Job-search automation for hh.ru — vacancies, responses, and resume boosting via Playwright.

View repository ↗
yandex-direct-mcp-plugin★ 12

Claude Code plugin for Yandex.Direct — MCP server + skills + OAuth.

View repository ↗
direct-cli★ 6

CLI for Yandex.Direct.

View repository ↗
claude_code_gmail_plugin★ 3

Gmail plugin for Claude Code.

View repository ↗
gemini-webapi-mcp★ 2

MCP server for Google Gemini — image generation, editing, and chat via browser cookies, no API keys.

View repository ↗
Case study tg_content_factory
Architecture diagram of tg_content_factory: Telegram sources feed a collector that stores messages in SQLite, surfaced through three search modes, keyword alerts and a FastAPI dashboard (diagram, not a screenshot).
The task
Keeping track of hundreds of Telegram channels and chats by hand does not scale: messages scroll away, built-in search is limited, and there is no way to watch keywords across accounts.
What was built
A self-hosted monitoring pipeline: multi-account collector with flood-wait rotation, SQLite (FTS5) message store, three search modes (local full-text, direct Telegram API, LLM agent), keyword alerts with bot notifications, anti-spam filters and a FastAPI web dashboard.
Verified result
Published on PyPI as tg-agent and shipped via tagged GitHub Releases (latest v0.2.3). Verifiable end-to-end scenario: pip install tg-agent, then python -m src.main serve brings up the dashboard on http://localhost:8080. No usage metrics are claimed.

Role Sole author — architecture, Telegram client layer, storage, dashboard.

Try it ↗ View repository ↗
Case study yandex-direct-mcp-plugin
Architecture diagram of yandex-direct-mcp-plugin: a Claude Code session drives the plugin, which exposes 147 MCP tools over the Yandex Direct API and installs from a plugin marketplace (diagram, not a screenshot).
The task
Routine Yandex.Direct campaign work means clicking through a slow web interface; an LLM agent can only help if the whole API is exposed to it as a structured, reliable tool surface.
What was built
A Claude Code plugin wrapping the Yandex.Direct API as 147 MCP tools — campaigns, ads, bids and budgets, keywords, reports, finance — with OAuth login through the browser, an ad-copywriting skill and a plugin-marketplace distribution channel.
Verified result
Installable from a public Claude Code plugin marketplace: /plugin marketplace add etopro/plugin-marketplace, then /plugin install yandex-direct@plugin-marketplace. Tagged releases up to v0.3.3; the full tool list is documented in docs/TOOLS.md in the repository.

Role Sole author — MCP server, tool surface, OAuth flow, skills.

Try it ↗ View repository ↗

Agent Tooling

claude-code-cycle-review-skill★ 7

Automated PR review cycle for Claude Code.

View repository ↗
clihost★ 5

Self-hosting / proxy CLI, security-hardened.

View repository ↗
zai-codex-helper★ 2

CLI to switch Codex between Z.ai and OpenAI (archived).

View repository ↗

Evaluation & Data

ccusage-dashboard★ 3

Hourly cost dashboard for Claude Code and Codex usage.

View repository ↗
llm_benchmark★ 3

LLM benchmark harness.

View repository ↗

03 Contributions

Useful changes, merged upstream

Every entry below links to accepted production code.

59merged upstream PRs

Featured examples — the full registry of merged PRs is listed below.

steipete / CodexBar ★ 21855Feature and documentation contributions to Peter Steinberger's desktop utility.Merged · PR #2814 ↗ yfinance ★ 25337Retry transient network errors.Merged · PR #2627 ↗ IBM / mcp-cli ★ 2021Unix-friendly automation and serialization.Merged · PR #227 ↗ agent-orchestrator ★ 12330Multiline submit and safer permission handling.Merged · PR #2357 ↗ ccusage ★ 18725ZCode SQLite usage adapter and report/test integration.Merged · PR #1675 ↗
All 59 merged pull requests

04 Experience

Built, shipped, maintained

01

Open-Source Contributor

2024 — Present

CodexBar · yfinance · IBM/mcp-cli · agent-orchestrator · ccusage

  • 59 merged PRs across upstream projects.
  • Reliability fixes, production features, automation, and safer interfaces.
  • Changes are scoped, tested, reviewed, and maintained after merge.
02

Independent Python Engineer

2021 — Present

Telegram automation · AI agent tooling · data & finance

  • Production services built around Python, asyncio, agents, and APIs.
  • Author of 16 starred open-source projects.
  • 136 stars across own projects and maintained forks.

05 About

Calm systems for complicated work

I build AI agent tooling, Telegram automation, and data systems with a bias toward explicit behaviour, observable failures, and uneventful operations.

When a dependency needs a fix, I prefer contributing the improvement upstream instead of carrying a permanent local patch.

Working stack

  • Python
  • asyncio
  • aiogram
  • LangChain
  • MCP
  • pandas
  • statsmodels
  • sktime
  • LightGBM
  • Docker
  • Git
  • Linux

Open to contributions, tooling, and select contract work

Need a system that actually keeps working?

Got a difficult integration, unreliable workflow, or data process? I’ll help turn it into explicit, tested software.

Telegram ↗ GitHub ↗