I build human-driven AI swarms and sovereign infrastructure. Since 1998 — web, backend, embedded, now multi-agent.
#OpenToWork — looking for an outpost
“As always: three to five years too early.”
— Dilles
Python, TypeScript, multi-agent systems, headless architectures. From prototype to own server stack with Cloudflare, Nextcloud, local GPU inference.
Generative systems, diffusion workflows, realtime image generation, prompt architecture. Tool becomes medium.
ESP32, Arduino, RP2040. Sensors, actuators, embedded AI. Things that react, measure, behave.
From CAD to resin, enclosures for electronics, sculptural experiments, functional prototypes.
Spaces where code, hardware and room come together. Interactive works between art and experiment.
Questions that don't disappear: autonomy, perception, language — what machines don't understand, and maybe we don't either.
The second line of coordination — local devices, a peer mesh reaching about 10 km and resilience centres, for when power, mobile networks or cloud services go down. Everything with a “Logpy:” prefix is a part of it. Stopped in 2026 — no funding, no partners. The idea was good; I’m a developer, not a salesman. logpy.com
Phone agent for a doctor’s practice — calls are transcribed and sorted locally, nothing leaves the practice. Runs as a prototype, not public yet. Demo on request: [email protected]
Multi-agent AI system on macOS. Orchestrates Ollama, Qdrant, Redis, Playwright and ComfyUI through a NiceGUI/FastAPI interface. Custom agents with tools, memory and coordinated tasks. I still use it daily; development is paused — too expensive and too time-consuming right now.
Direct communication with visitors at reception — ALICE answers questions about the site and projects, forwards to a human when needed (Telegram). Currently primarily a honeypot: we observe what AI agents, scrapers and bots actually want. Stack small and cheap — local Ollama (qwen3.5) on own hardware, plus free-tier models (Llama, GLM, gpt-oss). Cost: zero. A toy for now — not a product.
Machine Assisted Reasoning + Tactical Intelligence Network — my personal assistant. Context-sensitive memory, prioritization, decision support. Runs locally on own hardware. Not for clients, not for sale — the tool I use myself. Development currently paused.
Write directly — lands with me.