I’m Lars, a Master’s student in Information Systems at the Technical University of Munich. I enjoy building systems that make sense of data, automate solutions and help people work more efficiently. 

Before TUM, I completed my Bachelor’s in Information Systems at the University of Münster, where I graduated top of my class and built foundational understanding of statistics, Machine Learning and the complexity of applying Systems to Organizations.

In my Bachelor Thesis, I combined my interests in Machine Learning and AI with a business case by Viadee Unternehmensberatung GmbH and was able to improve an LLMs ability to classify invoices by over 40%.

I'm curious about how AI and data infrastructure evolve together and how we can build tools that people really use. That curiosity drives most of my projects, from small prototypes to production systems.

At OMMAX, I worked on building data platforms and analytical models for private equity clients. I helped design reporting pipelines that supported portfolio-wide decision-making and created prototypes for generative AI use cases. One of my favorite projects was developing a revenue forecasting model for a large UK retailer’s store expansion strategy.

During my time at viadee, I explored how large language models can extract and classify information from unstructured text. I build a retrieval-augmented generation system for consulting contracts and achieved more than 80% accuracy on real data. 

This work also formed the basis of my bachelor's thesis, where I improved LLM classification performance by applying and evaluating automatic prompt optimization frameworks.

I engineered a data pipeline, preparing 378 real invoices for classification and was able to improve real world performance of the classification by 45%, lifting the F1 Score from 0.67 to 0.97.

At zeb, I worked on the DevOps and infrastructure side of a large cloud application. I helped automate deployments using Terraform and improved security in a Kubernetes environment.

My main project focused on certificate rotation within a PKI of the Kubernetes cluster where I utilized cross-signed certificates to enable rolling restarts and reduce the downtime for certificate rotation to zero.

My first industry experience was in SAP development at best practice consulting AG. I built an email notification system in ABAP that helped teams react faster to updates in data pipelines. It tought me how structured enterprise systems work behind the scenes.

Outside tech, I lead multi-day trail rides at Reiterhof Hirschberg. Guiding groups through long tours means planning routes, managing logistics, caring for horses and staying calm when things don't go as expected.

It's a good reminder that leadership is mostly about attention and care.

At TUM, I am currently focusing my studies on Machine Learning and Artificial Intelligence subjects.

During my semester abroad in Australia, I tried out something new and studied Psychology, consumer behavior and microeconomics. 

These courses helped me feed my curiousity about how and why people think and act the way they do.

In Münster, I got to take fantastic courses teaching me the fundamentals of all things Data, Information Systems and Business. 

During my time there, I took part in multiple international projects, such as a Design Thinking project in Atlanta Georgia in 2023 and visiting the German American Conference at Harvard Kennedy School in 2024.

For my academic achievements in Münster, I was awarded with multiple awards, graduating top of my class in 2025.

For my bachelor’s thesis, I explored how large language models can automate financial account assignments in invoices, a task that usually requires manual review.

I built a framework that tested different prompt optimization techniques such as DSPy and TextGrad.

By fine-tuning both prompts and evaluation loops, I increased the F1-score from 0.67 to 0.97 on real-world invoice data.

The project showed me how careful engineering and evaluation can turn theoretical AI concepts into something that actually works in practice.

Vocabmate is a side project I started to make my own language learning more efficient. It creates Anki-compatible flashcards automatically from any text, powered by a large language model.

I built the backend in Node.js (TypeScript) and deployed it on AWS ECS using Terraform for infrastructure management, while the Flutter web frontend runs on AWS Amplify.

What began as a weekend prototype has grown into a full web app.

This project allowed me to experiment with modern deployment pipelines and the real-world behavior of LLMs outside a research setting.

I have since shut it down due to AWS costs. You can still visit the website at vocabmate.com!

During an international project in Atlanta, I worked with students from Georgia and Germany to improve Southwire’s order-to-cash process.

Using Celonis process mining, we identified bottlenecks and designed a new workflow that reduced delays in order handling.

I was responsible for structuring the data analysis and designing the final presentation, which we delivered to company executives.

The experience taught me how collaborative design thinking and data analysis can complement each other, and how cultural diversity can make a project stronger.