Open source, and what I build for myself
The personal side: tools I needed and published, products I run at home, a game turned into an engineering project. Everything here is on GitHub under my own name. Client work stays out, except the two anonymized cases at the bottom.
WhatsApp MCP
An MCP server that lets an assistant read and send WhatsApp through a linked device. I run it on my own server for reminders, archives and a voice relay; the fork adds mentions, revoke, media and responder toggles.
On GitHub → Claude Code skill · open sourceManual
One command that turns any codebase into a designed field manual: architecture, how it runs, how to change it, as an artifact page, Markdown and an optional standalone HTML or PDF. Built because every handover needs one.
On GitHub → Reference pattern · TypeScriptClaude Telegram Relay
The smallest honest way to run Claude Code as an always-on Telegram bot: one daemon, one chat, the full tool set, with the setup for Linux, macOS and Windows. The pattern behind my own manager agent.
On GitHub → Own product · Laravel, AndroidInventory
A private inventory for several households: a Laravel backend with an API and an MCP server, and a Kotlin Android client. Scan, count, share a list, ask the assistant where the spare keys are.
On GitHub → Tool · PythonWhatsApp Analyser
Reads a WhatsApp export and answers the questions you actually have: who talks when, how fast people reply, what a group is really about. Pandas, no cloud, your data stays on your disk.
On GitHub → Game automation · JavaScriptBitburner scripts
A game about hacking, turned into an engineering project: scripts synced into Bitburner over its remote file API, a scheduler that spreads work over every server it owns, and a bridge so the editor on my laptop is the editor in the game.
On GitHub → Front-end mockup · HTML, JavaScriptPrompt
A static mockup of an AI chat interface: sidebar, thread, syntax-highlighted code blocks, no model behind it. The starting point I use when a client wants to see what an assistant on their data would look like before anything is built.
On GitHub → This site · Laravel packagesplotnikov.dev
This website as a Laravel library: one shell, one stylesheet, one script, a terminal panel with typed scenes, a printable CV, a contact form, tests for every page, deployed from main through a host app on one small server.
On GitHub →Two pieces of work, told the way they went
The client names stay out; the numbers are real. The full case studies live on scuttle.dev, my company site.
Modernizing a monolith: Phalcon to Laravel
The mess. A client was trapped in a legacy Phalcon system on an unsupported PHP, unable to scale or to find specialized talent.
The solution. A custom PHP middleware layer that lets both frameworks run side by side, then an incremental migration (the Strangler Fig pattern), module by module.
The impact. 60 percent lower latency, PHP 7.4 to 8.4 back in support, and the hotels that run on the platform kept working through the whole migration.
AI agents in the sales stack
The mess. Manual research and data entry were stalling a sales operation: hours a week of looking things up and typing them over.
The solution. LLM agents integrated into the existing workflow: they research prospects, draft in the sender's words and queue for a human tap. Evaluation set, cost cap and a switch-off in the runbook.
The impact. 16 hours a week of manual work reclaimed; nothing leaves the building without a person approving it.
Want to build on one of these?
Issues and pull requests are welcome on every repository. For work on your own system, my company site has the services and prices.