Backend engineer with a long PHP/Symfony background, active in the PHP community — now building self-hosted, privacy-first software in Go.
I ship and operate Svoi, an open-source personal AI assistant, and build local-first desktop tools for AI creativity: Partscaster for music production and Imagecaster for image generation. Everything leans local-first: no cloud lock-in, no subscriptions, full control over your data.
| Layer | Technologies |
|---|---|
| Backend | Go, PHP (Symfony), PostgreSQL, MySQL, Redis |
| Frontend | Vue 3, TypeScript |
| AI / ML | Local LLMs, RAG, Stable Diffusion, vision models, music generation |
| Desktop | Wails v2, Python, SQLite |
| Infra | Docker, nginx, GitHub Actions, Linux |
A generated track is a draft. Partscaster turns it into something that sounds like a band played it: swaps the instruments for real ones, has a guitarist play the song's chords and a drummer play fills, and mixes the track the way real records are mixed. A desktop app plus an MCP server for AI agents.
- Stack: Go + Wails v2, stem separation, sample-based drum & guitar replay, 106 genre presets
- Built on: the YuE2 music model ecosystem — Partscaster starts where generation stops
A self-hosted, open-source personal AI assistant. Streaming chat with agentic tools, long-term memory, a RAG knowledge base, custom assistants and templates, tasks and reminders, and a proactive scheduler. AGPL-3.0; the operator configures a single LLM provider (local Ollama or any OpenAI-compatible cloud) that serves every user.
- Stack: Go + Vue 3 + PostgreSQL/pgvector, multi-provider LLM chain with failover, SearXNG web search, admin panel, PWA
- Design notes: ARCHITECTURE.md
Desktop companion for Svoi's external knowledge base ("My PC"). It indexes local folders and uploads them as ciphertext over an end-to-end-encrypted blind relay — the server holds no decryption keys.
A privacy-first desktop app for local Stable Diffusion + LLM workflows. Pick a style, describe the scene in plain language, get the image — an LLM writes the SD prompt, the technical settings live in the style. All processing happens on-device.
- Instruction-based editing of real photos; mask-drawn smart remove with LLM context analysis
- Multi-step pipelines, ESRGAN quality output, session management, kids mode
- Server companion: remote SD/LLM backend with a one-click model catalog
- Building Partscaster and Imagecaster; operating Svoi (svoi-ai.ru) as a production, self-hostable product
- Working as a staff-level PHP/Go architect on a large legacy-modernization platform
- Architecture over syntax — document why, not just how.
- AI as a multiplier — LLMs handle boilerplate and research; design, error handling, and UX stay in human hands.
- Local-first and private — build software people can actually run and own.
- Ship, measure, iterate — working software over perfect abstractions.
Feedback and architectural discussion welcome: dmitry.samotoy@gmail.com · samotoy.ru