3 SYSTEMS LIVE · OPERATING

I build demand engines that run on agents.

20+ years in B2B marketing. Today I design, build and operate agentic systems — n8n, OpenClaw, Antigravity, LangGraph — that produce pipeline output at the cadence of a full team.

15 workflows built10 in production3 live systems< €25/month total infrastructure cost
CASE 01 · IN PRODUCTION

Agentic ABM  Engine:
Account-Based Marketing 
applied to my own job search

When I started my job search, instead of chasing poorly matched opportunities, I built an account-based marketing engine orchestrated through HubSpot and agentic AI.

~350 openings scanned weekly (5 rotating queries daily, Google Jobs via SerpAPI)
Automation workflow — job scanning and scoring pipeline
→ multi-layer disqualification: role-type exclusions, company blocklist, ~30-municipality geographic ICP
→ LLM fit-scoring against my profile, structured output per opening: score, rationale, strengths, gaps
→ hard 85+ fit threshold threshold between each role and my profile — below it, nothing passes
→ deduplicated against Notion
→ 5–7 hyper-personalized job applications submitted per week
System 01 workflow — autonomous job matching engine
→ human-in-the-loop approval before submission
Automation workflow — resume and cover letter processing
→ applications managed as deals in a custom HubSpot pipeline, with every recruiter and interviewer interaction tracked as CRM activity
Custom HubSpot Deals pipeline tracking job applications
THE DETAIL THAT MATTERS Some days the engine delivers zero qualified openings. That's the threshold doing its job. No junk enters the pipeline — the same disqualification discipline I'd run on any MQL funnel.
n8n · SerpAPI · OpenRouter · Notion API · Google Sheets · Playwright
CASE 02 · IN BUILD (2 OF 5 STAGES LIVE)

Content-to-Demand Engine

A German-language B2B content pipeline for agentic AI topics — designed to run at team cadence with an editorial staff of one.

LIVE TODAY
Automated topic discovery with dual scoring — a heuristic viral score plus an independent LLM evaluation
System 02 workflow — AI research and intelligence engine
→ script generation through a self-hosted LangGraph service, with structured output and automatic discard logic: scripts that don't meet the bar are rejected with a documented reason, never published
→ routed to Notion for human review, status synced back to the research sheet
Automation workflow — Notion review and status sync
IN DEVELOPMENT: visual production, publishing automation, analytics feedback loop.
WHY IT'S ON THIS PAGE UNFINISHED Because that's what operating a demand engine looks like. Architecture first, quality gates at every stage, then scale — and honest labels on what's live versus what's next.
n8n · LangGraph · LangSmith · Qdrant · OpenRouter (DeepSeek for volume, Claude for scripts) · Notion
CASE 03 · FOUNDATION

Self-hosted infrastructure — full data control, agentic stack under €21,50/month.

Everything above runs on infrastructure I built and operate myself. Self-hosted n8n behind a Cloudflare Tunnel, connected to Claude via MCP. Multi-model routing through direct APIs and OpenRouter — Claude for quality-critical output, GPT-4o-mini for volume tasks, DeepSeek for cost efficiency. Automated site deployment via Wrangler CLI. 

I build AI agents that turn repetitive work into infrastructure, leaving human judgment where it creates the most value. One example is my paperwork workflow: every incoming letter is scanned, OCR-processed, understood by an AI agent, and automatically renamed according to German business naming conventions.

Automation workflow — Notion review and status sync

Based on its understanding of each scanned letter, OpenClaw autonomously creates follow-up tasks, updates priorities, schedules the appropriate actions in my calendar, and keeps me in control through a Telegram chatbot, where I can review, adjust or override any decision before execution.

Automation workflow — Notion review and status sync


THE NUMBER THAT MATTERS The entire stack — 15 workflows, 3 systems, hosting, domains, tokens — costs less than €22 a month. Cost-per-output is a design decision, not an afterthought.
Ubuntu · n8n (self-hosted) · Cloudflare Tunnel / Pages / Workers · MCP · PM2 · ocrmypdf
JUDGMENT

What I keep, what I drop

Tools don't make the demand engine. Judgment about tools does. A few decisions behind these systems, and why:

01 DeepSeek for volume, Claude for words. Topic scoring and classification run on DeepSeek via OpenRouter at a fraction of the cost; script and copy generation runs on Claude, where quality is the bottleneck. Routing by task beats loyalty to any single model.
02 Hard thresholds over soft judgment. The 85+ fit gate and the script discard logic are code, not mood. When output volume drops, I widen discovery — I never lower the bar.
03 Human review on every shipped asset. AI drafts at scale; I approve what carries my name. The same pattern I'd enforce for any brand operating in a regulated market.
04 Self-hosted where it counts. Data flows through my own infrastructure, not a chain of third-party SaaS tools. Control and cost, both by design.
05 Continuous learning runs on its own pipeline. Specialized literature digitized into NotebookLM, insights connected across notebooks, feeding into my workflows — so the systems I build draw on current, structured domain knowledge, not just training data.
BACKGROUND
Experience

20+ years across B2B SaaS, e-commerce, consumer goods and retail — Samsung Electronics, New Balance, Tommy Hilfiger, Henkel. Growth marketing, demand generation, marketing automation and CRM since 2018 in Germany. MBA (Marketing), Cardiff University. German citizen · German, English, Spanish.

Certified stack

HubSpot (Marketing, Sales, Service, Reporting — renewed annually for 5 years) · Salesforce Sales Operations · Tableau BI Analyst · Meta Digital Marketing · Google Digital Marketing & E-Commerce · IBM Generative AI for Growth Marketing + Python for Data Science & AI

Want this pointed at your pipeline?

These systems currently run on a market of one. Their next assignment is your demand funnel.

Book a call
Scheduled through HubSpot — naturally.