I lead one of five teams building the in-house student ERP of Germany's largest university. We own three domains and five services, and I'm the one people ask how they work and why — from the ministry rule to the business process to the domain model. Alongside, I build agents that triage production bugs and prepare the fix.

• Technical leadership

  Tech lead of one of five teams on a university ERP: owns three domains, sets the bar, mentored 10+ engineers.

• Software architecture

  Domain-driven services on Kafka that stayed changeable through five years of shifting ministry rules.

• Agentic engineering

  Agents that triage production bugs and propose fixes at work; a lights-out agent factory at home.

1. 2021 – today

   IU International University · Munich›

   1. 2024Tech Lead
   2. 2021Full-Stack Developer

   In-house student ERP of Germany's largest university (~150k students), five teams: NestJS, Kafka, MySQL, Vue, AWS.

   • Domain owner. Tech lead of one of five teams; owner of and go-to person for 3 domains and 5 services.
   • Ministry rules. Guided ministry-driven rule changes into business processes, then into domain models.
   • Student lifecycle. Leave of absence, programme change, cancellation and their interplay, modelled in DDD.
   • Integration. Kafka + transactional outbox; SharePoint via his DAM service, Salesforce via the CRM bridge.
   • NestJS foundation. Custom decorators, interceptors, guards; an ESLint plugin encoding domain rules; 300+ specs.
   • Continuous improvement. Node 24, NestJS 11, Datadog monitoring, faster local dev — many small iterations.
   • Mentoring. Onboarded 10+ engineers and taught them DDD, NestJS and Kafka — in pairing, review, refinement.
   • Bug-triage agent. Watches Datadog, classifies bugs, drafts root causes, opens fix MRs after human sign-off.
   • Agentic setup. Spec-driven Claude Code; 7 MCP servers give the team's agents Datadog, Kafka, DB and Jira.
   • Micro-frontends. Vue micro-frontends for booking and the academic profile: leave-of-absence flows, programme-change checks, the original statistics view.
2. 2018 – 2020

   SquareTrade · San Francisco›

   1. 2018Full-Stack Data Engineer

   Device protection plans sold through retailers. ETL ingestion and analytics: Spring Boot, Informatica, Python, PostgreSQL.

   • Merchant onboarding. Integrated new retail partners into the ingestion pipeline for protection plans.
   • Merchant access. Contributed to the IAM service that let merchants manage their own users' access.
   • Device analytics. Captured anonymous device telemetry that fed customer profiling and dynamic pricing.
3. 2015 – 2018

   Edelweiss72 & Sheraptec · Munich›

   1. 2015Web Developer, part-time

   Web agencies, alongside studies — TYPO3 and PHP sites for clients including dm, SWM and Deloitte.

Over 5,200 commits across 30+ repositories in the last twelve months — products, infrastructure and the agent tooling I work with every day.

• Agent factory2026My agentic engineering setup. Issues go in, verified fixes come out — triaged, implemented, reviewed, deployed and checked by agents, with a human gate where the risk is.Claude Code · MCP · Bun · PythonWarden is the control plane: it turns a signal — a failing check, an error spike, a new issue — into an investigation, a verdict, an implementation, a review and a merge, then verifies the fix in production. Around it: Hermes, a 24/7 assistant in Slack; agent-gateway, which offloads checks, reviews and implementation to cheaper models; gateways for research, speech, email and image generation; and model choice from daily benchmarks, not habit. It runs on a Mac mini, a homelab and a VPS.Open →
• BasaltUI2026A React design system optimised for agents — themed charts, app shell and a streaming agent chat.React · Mantine · visxMantine v9 plus visx: themed charts, an app shell, data tables, forms and a streaming agent chat — each module documented so an agent composes it right on the first try. On npm with about 8k downloads a year, and the design system behind my own dashboards.Visit ↗
• FreePlanningPoker2023Real-time planning poker for agile teams — 10k users and 170k estimates, free and without signup.Next.js · WebSocket · PostgresEveryone estimates at once, hidden until the reveal, so the first number can't anchor the room. Real-time rooms over WebSocket, no accounts, no setup — 10k users and 170k estimates since 2023.Visit ↗
• RollHook2026Zero-downtime rolling deployments for Docker Compose, triggered by webhook.Go · TypeScript · Docker ComposePush an image, call the webhook or the GitHub Action, and RollHook rolls the Compose service one container at a time behind health checks — no dropped requests. It deploys this site.Visit ↗
• WeatherOrb2026One forecast for sky and sea — 35 weather and 8 marine models, blended, calibrated and drawn on a WebGL map.Python · TypeScript · WebGLBuilt because Windy shows you the models and leaves the judging to you. WeatherOrb syncs 35 atmospheric and 8 marine Open-Meteo models, calibrates them against real observations from airports, buoys and the DWD, and serves one forecast as a point API and a rendered map.Visit ↗

• sy-serendipityWebsite for a charter sailing yacht
• ntfy-macForwards ntfy notifications to macOS

1. 2015 – 2020

   B.Sc. Economics & Computer Science

   Munich University of Applied Sciences

2. Outside work

   Powerlifting, hiking, skiing, surfing — and long walks with my dog.