AI Engineer · IT Consultant · Dresden, Germany
I build AI agents that run in production — not just prototypes.
From LLM orchestration and fullstack web apps to secure infrastructure, backed by years of enterprise IT.
- 🤖 AI agents in production — Design, build, and operate autonomous agents with semantic memory, multi-LLM routing, and real-time monitoring
- ⚡ Fullstack web applications — Next.js / React / TypeScript frontend, Python / FastAPI backend, Supabase / PostgreSQL data layer
- 🧠 AI workflow architecture — Context engineering, prompt design, and structured agentic workflows that deliver reproducible results
- 🔄 Automation & data pipelines — Web scraping, NLP preprocessing, n8n workflows, Docker deployments, CI/CD pipelines
- 🏢 Enterprise governance — Security-first architecture and operational discipline from years of Microsoft Cloud consulting
- 🦞 Vega — Autonomous AI assistant on OpenClaw — researches, delegates tasks, and audits security 24/7 via Telegram on a network-isolated VPS
- 📡 project-beat — Freelance project discovery for the German IT market — scrapes 5 platforms 2× daily, deduplicates via fuzzy + embedding similarity, ranks by relevance (v1.2 live)
- 🖥️ Mission-Control — Single dashboard for agent state, task queues, server health, and deployments — one tab instead of five terminals
- 🛡️ ShieldClaw — Open-source prompt injection defense — 71 patterns, 4 active hooks, zero token overhead. Protects agents from adversarial inputs in production (MIT)
- 🔧 OpenClaw Skills — Composable agent capabilities for the OpenClaw ecosystem — including ShieldClaw (71 patterns, MIT)
- 🔨 Schliff — Deterministic quality scorer for AI agent instruction files — scores SKILL.md, CLAUDE.md, .cursorrules, AGENTS.md across 8 dimensions including security. Multi-format, anti-gaming detection, zero dependencies. 1017 tests (v7.1.1, MIT, PyPI)
- modelcontextprotocol/servers#3733 — Added a root
CLAUDE.mdfor Anthropic's official MCP reference monorepo (7 servers across TypeScript and Python). Merged April 2026 by @cliffhall. Self-audit with schliff returned 59.2/100 at 40% coverage — full walkthrough →
| Tool | Role |
|---|---|
| Ghostty | Custom-configured terminal with splitscreen layout |
| lazygit | TUI Git client — keyboard-driven version control |
| Wispr Flow | Voice-to-text AI — speak commands, code and prompts |
| Claude Code | Agentic IDE running in the adjacent pane |
| VS Code / Cursor / Antigravity | IDEs for visual editing, debugging, and AI-assisted coding |
| Git Worktrees | Parallel branch development without context switching |
Claude Code Ecosystem
My agentic development workflow runs on Claude Code with a curated set of plugins, frameworks, and custom skills — from structured planning to autonomous code review and security enforcement.
Official Plugins
Community Frameworks
Custom Skills
Tech Stack
AI & Agents
Dev Stack
Infrastructure & Deployment
Tools & Integrations
| Repo | Description |
|---|---|
| 🐍 hydra | Multi-headed cross-model code review — 6 AI advisors (Opus + Codex), 3 peer reviewers, chairman synthesis. Up to 10 agents in deep mode. Based on Karpathy's LLM Council. Claude Code skill (MIT) |
| 🔨 schliff | Deterministic quality scorer for AI agent instruction files — multi-format (SKILL.md, CLAUDE.md, .cursorrules, AGENTS.md), 8-dimension scoring with security, compare, suggest, anti-gaming. 1017 tests, self-score 99.0 [S] (v7.1.1, MIT) |
| 🌐 fpaul.dev | Personal developer website — Next.js 16, MDX blog, ASCII art animations, light/dark theming, Clawd mascot. Static export on Vercel |
| 🛡️ openclaw-skill-shieldclaw | Prompt injection defense for AI agents — 71 adversarial patterns, zero token overhead, 133 tests, MIT licensed |
💬 Interested in working together? Let's connect on LinkedIn
I spent years in Microsoft Cloud consulting — deploying M365 environments, managing tenant security, and building automation for enterprise clients. That experience shapes how I approach AI engineering: production-grade infrastructure, security by default, and operational discipline over hype.
More recently, I've built automated data pipelines with German-language NLP — compound splitting, semantic matching, and embedding-based deduplication for production use cases.
Today I build agents that run unattended in production, not demos that break after the first edge case. The enterprise mindset is the differentiator — I think in terms of uptime, audit trails, and access control, not just prompt quality.
Available for freelance and consulting projects in AI agent development, fullstack web applications, NLP/ML systems, and automation.