I am a physicist turned AI/ML specialist with 12 years in the field, including five years of research and seven years of consulting experience.
I combine experimentation and theory with a pragmatic approach to business. I aim to bridge cutting-edge AI and real business value — staying on top of the latest developments, but always going for the “good-enough” implementation.
I currently head Netcompany’s AI Task Force, specializing in AI and ML advisory work and early-stage development. I also lead the AI center of excellence at Netcompany.
“A pioneer in his field with a uniquely deep subject expertise due to his research background” - Gustaf Löfberg, CDO of Netcompany
“Although we regularly speak with some of the most skilled AI experts in Denmark, I have yet to meet anyone who can do what Mathias can” - Christian Morten Prip, principal at Netcompany
“It was immediately clear that Mathias has an extraordinarily sharp and analytical mind” - Morten Scheibye-Knudsen, professor at University of Copenhagen
During the pandemic, the National Health Service was fighting to improve better uptake of vaccines in the UK population. People booked for a shot, but did not show up. My team was put to the task of analyzing data and building a machine-learning model to predict the no-shows. This enabled the NHS to better target their efforts in public and individual outreach.
This task was of course highly motivating, though extremely challenging. Not least due to the tight technical restrictions on working with the entire UK’s health data on an individual basis. Further, the ML task was hard, due to an extremely unbalanced dataset between shows and no-shows. The NHS was able to use our highly accurate model predictions to better their efforts at the second wave of the pandemic.
DSB, the national railway operator in Denmark, had a problem. Passengers are actually okay with minor train delays, but they hate not knowing when the train arrives. DSB engaged my team and me to develop an AI system to predict train departures and arrivals on a national scale. This project ran from very early experiments on single lines, to MVP-deployment in production, to full-scale Databricks/Pulse automated MLOps.
I grew the project over 7 months from a team of 3 to a team of 30. At national scale, the system broadcasts predictions in real-time across Google Maps, Journeyplanner apps, web and physical screens all around Denmark. The core architecture I developed is based on a swarm of many models that makes roll-out, debugging, retraining and maintenance easier than a single neural architecture.
Taking a project from idea to full-scale multi-model production system is extremely rewarding and still rare. A leading mobility provider was in need of real-time market predictions and a similarity search system. Immediate predictions allowed them to trade with precision beyond each trader’s intuition in a rapidly changing market.
The platform supported fully automatic MLOps across environments for six different ML models. I was the delivery lead on this project, from training the first model myself, to supporting a larger team of data scientists and developers in later implementation. My most important learning was the importance of continuously connecting business end-users’ needs with the development - this assures adoption when you reach the end product.
Besides the three projects listed at the top, here are some selected cases.
The agency is creating nation wide communication and information hubs for danish citizens. Here I have contributed to GenAI PoCs, to explore GenAI in secure public citizen service. The work has centered around combining the public knowledge RAG systems with the digital personal post.
Architected a scalable Agentic RAG solution for a digital communications platform, translating complex business requirements into a technical scope. Designed a comprehensive evaluation framework to rigorously measure agent behaviour, ensuring reliability and safety in production environments.
Designed and deployed a parameter-free machine-learning model to identify atypical accounting entries. The model analyses metrics across seven years of financial statements for ~100 municipalities, scanning 12 million entries to uncover previous fraud across standard national accounting.
Technical strategy work for the company’s AI department. Design of the roadmap, architecture, and best-practice guidelines to support and streamline the development, operation, and production deployment of Machine Learning models (MLOps).
Designed and implemented a high-performance social economic statistical cost-benefit model. The model evaluates the economic impact of school-sector interventions across Danish municipalities. The model merges user-provided parameters with national statistics and scientific evidence.
AI advisory work and data exploration across the real estate management’s three core business streams: CAPEX, rent setting and customer service. The project scoped five GenAI and ML cases across ticket handling, data-driven CAPEX, predictive maintenance and dynamic pricing.
Designed AI solution that extracts and anonymises personally sensitive information and supplies domain-relevant keywords to users. The model runs multiple modern techniques in parallel (including local GenAI) and runs as a stateless micro-service integrated into a large case system.
AP Pension, Arla Foods, ATP, Bestseller, Copenhagen Infrastructure Partners, DSB, Digitaliseringsstyrelsen, Energistyrelsen, Forca, Grundfos, Gældsstyrelsen, Heimstaden, Copenhagen Municipality, Landbrugsstyrelsen, Lån og Spar, Munich Airport, Netcompany Banking Services, Network Rail, Odense Universitetshospital, Southeastern, Sparekassen Kronjylland, Styrelsen for patientsikkerhed, TopDanmark, Unifeeder, Velliv, Vy
Go Ahead Nordic, Nordea, Udvikling- og Forenklingsstyrelsen
Experienced in delivering multi-team IT projects. Focused on people development. Clear communication in English and Scandinavian languages.
Statistics, ML, AI, GenAI/LMMs, RAG, NLP, Data Engineering, Analytics, CI/CD, Data Insights
Python, Pandas, Rust, SQL, PHP (Backend), R, PowerBI, Bash and Pyspark
Claude Code, Databricks, Microsoft Azure, Git, Docker, Kubernetes, Pytest, GitHub, MLflow, SQL Server, DAX, Snowflake, Scikit-learn, Informatica, Azure ML, FastAPI and Dagster
Scrum Master ProfessionalTM, Microsoft MTA 98-364: Database Fundamentals, Teradata Vantage for Data Science using New SQL, Python and R
Written by hand. References on request.