Most of my projects start the same way: I run into something annoying, wonder why there isn't a better tool for it, and eventually build one.

I care about software that becomes useful quickly, feels like a product rather than a demo, and turns AI output into something people can actually use.

See where your attention actually went.

A local-first browser extension for reconstructing what you passed, noticed, clicked, and actually read while browsing. The first adapter focuses on Linux DO; the broader idea is personal analytics for attention without sending browsing history to a server.

Turn raw YouTube subtitles into a study guide.

A pipeline for transforming messy subtitles into structured bilingual learning material with timestamps, vocabulary, IPA, Markdown, and printable PDF output.

Turn AI Markdown into Word and PDF without wrecking the document.

A document pipeline for converting Markdown-heavy AI output — including equations and rich formatting — into files that are easier to submit, edit, and share.

Add Russian stress marks anywhere on the web, offline.

A Chrome/Edge extension for Russian learners that annotates stress directly on webpages, selections, and inputs using a fully local dictionary pipeline.

• 🎙️ gemini-tts-deck — a recording-console interface for AI speech generation.
• 📥 youtube-subtitle-downloader — a small agent-friendly tool for extracting structured subtitles.
• 🗺️ agent-atlas — experiments around agents, tools, and autonomous workflows.
• 🌐 laiekankan — another product experiment from the same “build the tool I wish existed” habit.

• Useful before impressive. A project should solve a real annoyance before it shows off a stack.
• The first 10 seconds matter. People should understand what a tool does without reading an architecture document.
• Local-first when it makes sense. Personal data should not need to leave the device just to become useful.
• AI should produce artifacts, not just answers. The interesting part often begins after the prompt.

After the Prompt — what happens when AI output has to become something people can actually use, edit, ship, or learn from.

I'm spending more time on agent-native tools, personal software, and making open-source projects easier to understand and launch.

If one of these projects is useful to you, open an issue, send a PR, or follow along — there will be more.