The machine
01Correlion OS is my own harness for running AI agents as a real team. Specialized roles work through explicit review gates, write gates, and CI/CD I designed. I hold the judgment and the final call; the agents carry the throughput.
Selected work
02Milan
A WhatsApp nutrition assistant for a longevity clinic.
Log a meal by text, voice note, or photo; an LLM pipeline extracts structured macros from any description. Built end to end on Azure over the WhatsApp Business API, with durable per-user data, prompt-injection hardening, and a QA dashboard scoring extraction quality meal by meal, under EU data-sovereignty controls for clinical data.
Berlin
The warehouse the whole company reports off.
A post-merger group built from two backends. I rearchitected the analytics warehouse into clean staging, intermediate, and core layers in dbt and BigQuery, advising the CEO, Head of Data, and Head of Supply directly. Fixed tens of thousands of duplicate records, shipped self-serve Looker reporting, and put a custom dbt runner on Cloud Run running daily production transforms. Multi-million-euro decisions now run on corrected data.
Co-founder & CTO
A legal-AI product, built solo, taken to funded.
I built the entire product alone: infrastructure, RAG pipeline, backend, frontend, and deployment. A legal document search tool with source-grounded answers and streaming UX, taken to an in-market product and 280,000 US dollars in pre-seed funding.
Engagements
03Technical leadership, part time: architecture, delivery, hiring, and the hard calls, for companies that need a CTO's judgment without the full-time seat.
Warehouses, pipelines, and reporting people trust. dbt, BigQuery, batch and streaming, from messy reality to a clean core the whole company can run on.
LLM products and features built end to end: retrieval, agents, evaluation, cost control, guardrails, shipped to production.
Agent systems that do real work under real gates, like the one that runs this company. I build them for yours.
The through line
04
I set out to understand human behavior. My bachelor's was sociology, political science, and economics, with biochemistry and neuroscience alongside for the biological half of the question. I wanted harder answers than the social sciences could give me, so I went to Cambridge on a Gates Scholarship: genetics, genomics, evolutionary psychology. Applied statistics ran through all of it, and statistics led me to data. Then I took the question into the field, evaluating multi-million euro government programs for Germany's federal ministries, weighing qualitative and quantitative evidence together. From there the path was business intelligence, data engineering, and data science at scale, then a startup of my own, and now Correlion.
Every system I build is another way of asking it.
Products
05Correlion Lab / Bet 01
Better Map
The first product out of Correlion's own lab, built and shipped by the same agent system that serves clients.
Before freelancing
06Senior Manager, Data Engineering and Senior Data Scientist at Gojek, running fraud and risk anomaly detection over 2M+ daily transactions. M.Phil. Biological Anthropology, University of Cambridge (Gates Scholar).