I build infrastructure for AI agents, graphics engines in Rust, and tools that bridge hardware and software.
Verdict — Guarded agent runtime for Rust
A comprehensive Rust framework for building and orchestrating guarded AI agents with safety guarantees, structured pipelines, and MCP integration.
I work across the stack — from GPU-accelerated path tracing and real-time fluid simulation to semantic memory systems for LLM agents. My work is driven by a conviction that the most interesting problems live at the boundaries: between hardware and software, between graphics and simulation, between AI and structured tooling.
I started in C# and Java (2018–2022), building desktop applications, audio hardware integrations, and graphics simulations. A deep dive into Rust began in 2023 and has since become my primary language — it now powers everything from WebGPU engines to MCP servers to CLI tooling.
AI agent infrastructure — building the memory and task management layer that lets LLM agents operate with persistence, context, and structure. My MCP servers provide semantic retrieval and hierarchical task orchestration for agentic workflows.
Self-hosted agentic CI — odev is a self-hosted runner that boots isolated Docker containers per task, streams a live opencode TUI over WebSocket, and surfaces diffs, file trees, and per-repo workspaces in a Leptos dashboard.
Graphics & simulation in Rust — a WebGPU engine with ECS-based scripting, a physically-based path tracer with Blender integration, and voxel terrain streaming at scale.
AI / LLM Infrastructure
odev(private) — Self-hosted agentic CI runner: isolated Docker sessions per task, live opencode TUI streamed via WebSocket, syntect-highlighted diff viewer, file browser, per-repo memory & task workspaces. Rust + axum 0.8 + Leptos 0.7 (SSR+WASM) + bollard 0.18. Single binary via cargo-leptos, deployed via Docker Compose atghcr.io/eliasstepanik/odev-server:latest.memory-mcp-server— Semantic memory for AI agents: hybrid FTS5 + vector retrieval, multi-provider embeddings (Voyage, OpenAI, Cohere, Ollama), temporal decay, cross-encoder reranking. Multi-arch Docker builds (amd64/arm64) → GHCR.task-mcp-server— Hierarchical task management with semantic search, importance scoring, and full CRUD for agentic workflows. Multi-arch Docker builds (amd64/arm64) → GHCR.Audiobook-Generator— LLM-powered multi-speaker audiobook pipeline using Qwen3 + TTS. 3-variant Docker builds (CUDA / CPU / ROCm) → GHCR.
Graphics & Game Dev (Rust)
|
|
|
|
|
|
|
|
|
Astronomy Automation
n8n-nodes-ninaapi— n8n workflow nodes for N.I.N.A. astronomy software: cameras, mounts, domes, focusers, polar alignment
Infrastructure & DevOps
IonosDDNSUpdater— Dockerized dynamic DNS updater for Ionos API. Dev/prod branch pipelines + .NET 6 build+test suite.AutoProxy— Automated proxy management infrastructure
CI/CD
memory-mcp-server+task-mcp-server— Multi-arch Docker builds (amd64/arm64) pushed to GHCR on every push to mainAudiobook-Generator— 3-variant Docker images (CUDA / CPU / ROCm) built and published to GHCRIonosDDNSUpdater— Separate dev/prod branch pipelines with .NET 6 build, test, and publishbevy-project-template— Full CI/CD suite: dependency audit, continuous integration, continuous deployment- Several others: ClusterManager, strudel-docker, WorldOfPadman_Dedi_Server, zmk-config, webgpu-engine
Audio & Hardware
|
|
|
Running on a self-managed homelab:
AI / LLM services — Ollama (local LLM inference), meridian (Claude proxy), audiobook-generator, qwen3-tts
MCP infrastructure — portainer-mcp, memory-mcp, task-mcp (custom MCP servers for AI agent tooling)
Data layer — Qdrant (vector database), SpacetimeDB, OpenCloud
Workflow automation — n8n
Networking — Headscale (WireGuard mesh), Tailscale agent, Caddy reverse proxy
Monitoring — Beszel, Watchtower, Skyserver
Game servers — Factorio, Terraria, Satisfactory, Project Zomboid, Enshrouded, Abiotic Factor
Personal infra — Syncthing, Protonmail Bridge, Pterodactyl
Astronomy — nova (DSO tracker)
| Channel | Address |
|---|---|
| Email (Personal) | eliasstepanik@proton.me |
| Email (Volkswagen) | elias.stepanik@volkswagen.de (Volkswagen-related work inquiries only) |








