Dimitri Rupp · Vienna · Infrastructure & AI collaboration
Every project in this portfolio runs in production, is verifiable via live status, and states precisely which work is my own and which was created in collaboration with Claude AI, Claude Code and Claude Cowork.
// every figure links to its source — no number without evidence
Multilingual AI news portal with a fully automated pipeline: 36 verified sources, 14 topic categories, RSSHub/FreshRSS aggregation, AI-assisted editing, daily briefing via n8n, Matomo analytics. Every statement is traceable to its original source — transparency in line with Art. 50 EU AI Act.
Self-hosted on Hetzner/Coolify, operated to enterprise standards. The bar shows the documented division of work: human ↔ AI.
Company website with an integrated, self-developed AI support assistant as a WordPress plugin — answers questions on Windows-to-Linux migration and products around the clock, incl. file upload.
Situational awareness dashboard for the outage control centre: GeoSphere weather warnings and outage data from 9 grid operators in one common picture, auto-refresh every 120 s — developed for SPL Tele.
Prometheus, Grafana and Blackbox Exporter across the entire fleet — with an alert relay that turns Grafana alerts into Zammad tickets automatically.
Interactive cost comparison tool: open-source enterprise stack vs. proprietary SaaS — and the reference for every market value shown in this portfolio.
Structured customer support with SLA capability — integrated with Level.io RMM and the Grafana alert relay, full data sovereignty in an EU data centre.
Self-built multi-workstation synchronisation for AI working environments: CLAUDE.md, skills and plugins kept consistent across three stations, incl. bootstrap for new devices.
AI-assisted classification service for Paperless-ngx: automatic document assignment via API with correspondent logic and alias mapping.
Infisical as programmatic secrets manager with machine identities plus MetaMCP as gateway — the security foundation for AI agent workflows.
Self-hosted password management with mandatory MFA and encrypted Restic backups on Hetzner Storage Box — restore verifiably tested.
Example: Zammad Helpdesk. Structure: problem → solution → result, complemented by full collaboration and verification documentation.
id: zammad-helpdesk · status: production · 2025-11 →
Dolibarr tickets did not scale for structured customer support with SLA requirements. Full-text search, escalation logic and multi-client capability were missing.
Self-hosted Zammad on Coolify: Traefik routing, Elasticsearch full-text search, Level.io RMM integration, Grafana alert relay and connection to the two-track Restic backup concept.
| Period | since November 2025, in production |
| Effort | — PD (Git history as evidence) |
| Stack | Zammad · PostgreSQL · Elasticsearch · Coolify · Traefik · Hetzner |
| ALEM zone | Zone 3 — Service Operations |
| Market value | — €/a Method: comparison against public SaaS list prices (reference linked) |
| Verification | Live status badge · weekly auto-screenshot · repo on request |
99 % of the work in this portfolio is created with Claude AI, Claude Cowork and Claude Code — in a fixed, disciplined workflow.
Claude AI and Claude Cowork for web research, source cross-checking and document work — organised in Claude Projects with project-specific instructions, following the evidence principle: no statement without a source.
Claude Code for agentic multi-file coding with Git discipline — governed via CLAUDE.md files and own skills, connected to external systems through self-built MCP integrations. Secrets are delivered programmatically from Infisical (machine identities); credentials never appear in plain text.
Every AI output is verified before it counts: BVL acceptance, monitoring via Prometheus/Grafana and multi-workstation governance via claude-workstation-sync.
Three principles make this portfolio verifiable rather than declarative.
No figure without a linked source. Market values carry their calculation method visibly: SaaS list price comparison (via swcalculator) or effort × market rate — never a bare number.
Every project states AI contribution and own work separately. Verifying AI outputs is always own work — the core of responsible AI collaboration.
Five criteria, no in-between: factually correct, enterprise-grade, externally auditable, standards-compliant, complete. Only 5×1 appears here.