haiodo.blog
Andrey Sobolev

Andrey Sobolev

whoami

Software engineer, more than twenty years in development. I write in anything that moves; if it doesn't move, I push it and write in it. I enjoy learning new technologies and sharing what I learn; there isn't always time for the sharing part, but it happens.

I've built up an understanding of the technologies, processes and standards that normal development can't happen without: enterprise software, project management, test automation, architecture. I like building tools people then use every day, and getting them to the point where they don't get in the way.

Novosibirsk State University: bachelor's (2001 - 2005) and master's (2005 - 2007) in information technology. For three years I've been teaching students object-oriented programming in C++ and Java. Russian is my native language, English is my working language.

cv.pdfcv-ats.pdfcv.md

support

Work on open-source projects often needs support. If you like what I do, please support me:TributeBoosty

history | tail

ls ~/projects

▸Intabia PlatformTypeScript · 2025 -

open platform for teams, what I work on now

An open platform and a single workspace for teams: Chat, Tracker, Office, Drive. Available on Web, Desktop, iOS, Android and through MCP.

platform.intabia.rugithub
▸Platform CollectiveTypeScript · 2026 -

the platform continued by the community

The same foundation as Huly, but development happens in the open, by the community. A fork that lives its own life.

Platform-Collectiveplatform
▸GOWTGo · 2026 -

native GUI for Go without cgo

A library for building Go applications with native widgets on macOS, Windows and Linux. The Go API follows Eclipse SWT, adapted from Java to Go. Plus clean Go approaches.

githubchangelogOn the benefits of generative AI, or how I built my own UI framework
▸Hum1izerGo · 2026 -

finds officialese and AI traces in prose, code comments and commits

A Go CLI: checks prose, code comments and commit messages for clichés, officialese and AI slop, reports findings with line numbers and a suggestion. Plugs into Claude Code, Codex, opencode and pi as a skill and hooks: the agent sees the finding right after an edit. It doesn't write text for you; the human decides.

githubhow it was builta week with hooks
▸OAITTPython / Swift · 2025 -

speech recognition on GigaAM with an OpenAI-compatible API

A speech recognition service on the GigaAM model with an OpenAI-compatible API. Two implementations with one API: a Python service and a native Swift/MLX build for Apple Silicon as a macOS menu bar app. Workers run as separate processes with auto-restart, live stats and benchmarks.

github
▸TenniarbGo / TypeScript · 2018 -

design tool with a text format

A personal project: a design tool where interfaces live in the .tenn text format — components, styles, parameters, animations and computations in nodes. A Go app for macOS, Windows and Linux, a web version and MCP for agents.

project overviewlive demogithub
▸OOPC++ / Java · 2023 -

OOP course and assignments for NSU

An object-oriented programming course and assignments that I teach to students at Novosibirsk State University.

github
▸HulyTypeScript · 2021 - 2025

open platform for business applications

An open platform for applications: Chat, Tracker, HRM, ATS. Tracker, GitHub Integration, system and UI components are my work. 27.9k stars on GitHub.

huly.iogithub
▸Network Service MeshGo · 2018 - 2020

hybrid/multi-cloud IP service mesh

A hybrid/multi-cloud IP service mesh for Kubernetes. I worked on the core and integrations.

networkservicemesh.iogithub
▸YANG-IDEJava · 2014 - 2016

IDE for the YANG / NETCONF modeling language

An IDE for the YANG language, developed at Xored for Cisco. Product, build, editors, model. The fork lives on in OpenDaylight.

github
▸ECLJava · 2010 - 2013

Eclipse Command Language, the scripting language of Q7 and RCPTT

An extensible command-line scripting language from Xored, the main scripting language of Q7 and RCPTT. I built the core, the debugger runtime, redesigned client-server command execution and added new commands.

github
▸F4Java / Fantom · 2010 - 2016

IDE for the Fantom language on Eclipse

An IDE for the Fantom language based on Eclipse and DLTK. JDT integration, completion and FFI navigation, parser, build; later I maintained the project and took community PRs.

github
▸Eclipse RCPTTJava · 2009 - 2021

UI testing tool for Eclipse applications

A record-and-replay UI testing tool for Eclipse/SWT applications. It grew out of the Q7 project at Xored and became an open Eclipse Foundation project. The ECL scripting language, distributed test execution.

eclipse.dev/rcpttgithub
▸RCPMLJava · 2006 - 2007

declarative UI in XML for Eclipse RCP

A UI engine for Eclipse RCP: the interface is described in XML and styled with CSS. I built the core, forms and SWT rendering.

github
▸Eclipse DLTKJava · 2005 - 2015

Dynamic Languages Toolkit, one of the first contributors

A framework for building IDEs for dynamic languages - TCL, Ruby, JavaScript. One of the first contributors to the project, I built the framework foundation and full IDEs on top of it.

eclipse.dev/dltkgithub

git log --graph platform

One foundation, three generations - from Huly to what I do now.

  1. Intabia Platform2025 -

    Intabia Fusion, I'm a co-founder. The product branch of the platform I work on today.

    intabia-fusion/platform - PRs
  2. Platform Collective2026 -

    The platform continued by the community. Same foundation, open development.

    Platform-CollectivePlatform-Collective/platform
  3. Huly2021 - 2025

    Huly Labs. An open platform for applications: Chat, Tracker, HRM, ATS. My work - Tracker, GitHub Integration, system and UI components.

    hcengineering/platform

head -5 posts/

The benefits of generative AI, or how I made my own UI frameworkread →

How my experience with Eclipse, a Java-to-Go translator, and working with agents helped me build gowt.

oQ in oMLX: what I found about Qwen3.8 quants and where my measurement disagreed with expectationsread →

A month ago I benchmarked engines for a local Qwen3.8-27B. Since then oMLX has improved its oQ quantizer, dozens of oQ builds of Qwen3.8 have appeared on Hugging Face, and in 0.7.1.dev1 oMLX learned to serve decision models. I collected what is known and re-measured on my M4 Max 48 GB: oQ4e against plain 4-bit, with and without MTP, and the clef decision model on 600 labeled URLs.

57% fewer long comments. My hook had almost nothing to do with itread →

A week ago I found that my comment rules never reached Claude's subagents, and fixed it. I re-measured on 11 thousand edits from Claude transcripts: the 57% drop came from a model change, the rules at startup changed almost nothing. What works is the other hook - the one that shows a finding right after the edit.

The em dash doesn't give away a neural network. I checked on a corpusread →

I measured almost everything that's considered a sign of machine text on 7500 texts. Half of the signs don't work, some work backwards, and my own detector gave the agent's texts higher scores than mine.

all posts →

contact

├─ email     haiodo@gmail.com
├─ email     haiodo@yandex.ru
├─ telegram  t.me/dev_haiodo
├─ linkedin  linkedin.com/in/sobolevandrey
└─ github    github.com/haiodo