About the project, about us 

We see a future where claim management runs on prediction, not surprise. Today, repair costs appear long after the damage is done — slowing down every decision. But the data to forecast them has been there all along. At claimini, we’re unlocking it. 

We manage vehicle claims for fleets — from the first damage report to the final repair — and we do it differently: digital, data-driven, and a little cooler than you’d expect from the insurance world. 

In this project, we want to build an explainable machine learning system that predicts repair costs from day one and keeps refining predictions as the claim progresses. The result: early cost control, better budgeting for fleets, and transparent communication that actually delights. 

You’ll develop a model using real (anonymized) claim data, train and evaluate it, and make it explainable. Together, we’ll visualize everything in a clean Streamlit dashboard that turns complex data into instant insight. We’re based in Hamburg’s city center — drop by anytime for good coffee, great data, and a genuinely fun team that believes Schadenmanagement can be both smart and relaxed. Ready to make claim management a little cooler?

What you’ll do 

You will build an explainable machine learning model that predicts repair costs from the moment a claim is created — and keeps refining those estimates as new data comes in. Starting with around 50–75 k anonymized claim records, you’ll engineer features, train regression models such as Random Forest, XGBoost, and LightGBM, and extend your predictions to fleet-level forecasts using time-series methods like Prophet or ARIMA. To make your results transparent, you’ll apply SHAP (and optionally LIME) to reveal which factors drive repair costs. Finally, you’ll bring it all to life in a Streamlit dashboard that visualizes real-time predictions, confidence ranges, and fleet-wide insights. 

We're looking for 3-4 master's students (data science, software engineering, machine learning) who: 

• Speak Python (pandas, scikit-learn, XGBoost/LightGBM)
• Have experience or strong interest in feature engineering, regression modelling, time-series forecasting, model explainability
• Are curious, motivated, and love turning real-world data into insights
• Want to work in a dynamic, business-driven environment

Apply to this project here