Ceraluna Labs · Lingen (Ems), Germany
Ceraluna Labs is the independent practice of Dr. Tina Koziol: computational economist by training, automation engineer by habit. Ten years of turning messy data into systems that keep running, from a 240,000-agent model of financial contagion to LLM pipelines reading corporate documents at 99% accuracy.
Services
Engagements range from short technical reviews to multi-month builds. Every project gets the same treatment: clear scope, measurable outcomes, and systems that keep running after the handover.
Agentic workflows, retrieval systems and LLM-powered data products, with the evaluation and governance work that makes them dependable. My last corporate LLM pipeline read messy, real-world documents at 99% extraction accuracy because it was engineered and measured, not just demoed. I have also led EU AI Act and NIST AI RMF alignment for an organisation of more than 120 people.
Predictive modelling and pipeline architecture on PostgreSQL and Python stacks, taken all the way to production. Four years as Senior Data Scientist at Wintershall Dea: LSTM anomaly detection on live well sensor data, model governance, and the CI/CD and monitoring layer that meant nothing stayed stuck in a notebook.
Agent-based models, simulation and policy analysis. My PhD put 240,000 agents on HPC clusters to study how financial crises spread through networks; the resulting work on systemic risk is peer reviewed and cited, including in the Journal of Economic Behavior & Organization. Current research covers monetary policy, housing and inflation.
Site migrations, structured data and programmatic content pipelines, handled as engineering problems rather than guesswork. capetowndata.com runs this way: nine locales on Django with automated translation, keyword-driven report generation and scheduled data pipelines, served through Cloudflare.
Selected work
A programmatic data property about Cape Town in nine locales: Django, CI/CD via GitHub Actions, automated content translation, crime data pipelines, interactive maps and scheduled report generation. Designed, built and operated end to end by one person.
An independent iOS app on the App Store. Concept, data modelling, development and App Store release handled solo, from first commit to shipped product.
Journal of Economic Behavior & Organization, 2023. A network SEIR model in which agents learn about the disease from their neighbours, calibrated to Cape Town: social learning both flattens and shortens the epidemic curve. Current working papers examine monetary policy, housing and asset bubbles.
About
Tina Koziol is a computational economist and data scientist based in Lingen (Ems), Germany. For her PhD at the University of Cape Town (2020) she put 240,000 agents on high-performance computing clusters to study how financial crises spread; the resulting work on systemic risk and contagion is peer reviewed and cited, including a 2023 article in the Journal of Economic Behavior & Organization.
She then spent four years as Senior Data Scientist at Wintershall Dea in Hamburg, taking AI out of notebooks and into operations: an LLM document pipeline with 99% extraction accuracy, LSTM-based monitoring of live well sensor data, the MLOps layer underneath it all, and EU AI Act alignment across more than 120 staff.
Since 2025 she runs Ceraluna Labs, building automated data platforms end to end and continuing her research on monetary policy, housing and financial markets. Her own products include capetowndata.com and timeline.music. She works in English, German and French.
Contact
If you have a project in economics, data science or applied AI, send a short note about what you are trying to do and where it currently stands. You will get a direct, honest reply from the person who would do the work, including a referral elsewhere if it is not a good fit.
Based in Lingen (Ems), Germany. Working remotely with clients in English, German or French.