Jurisdiction - AI entity labeling tool for Dutch legal rulings

Tech Stack

Overview

Rechtspraak is an AI/NLP project developed for Arbeidsmarktresearch (University of Amsterdam). The goal is to improve the readability of Dutch court rulings by automatically labeling key entities in legal text.

Court rulings are often long and written in dense language, making it difficult to quickly scan who is speaking and what each part is about. This project helps by highlighting important roles and concepts inside the document.

My role (6 months)

This was a full-semester project where I worked on almost the entire pipeline:

• Research and documentation
• Data preparation and processing
• Model development and training (NER)
• Improving the document workflow and output format
• Supporting the end-to-end application flow
• Writing and delivering the final report (Dutch)

Project goal

Automatically label legal text with useful entity types such as:

• Parties (e.g., werknemer / werkgever roles)
• Judges
• Legal topics / concepts

This makes rulings easier to review and helps users quickly understand the structure of a case.

AI/NLP approach

The core technique used in this project is Named Entity Recognition (NER), where a model predicts entity labels on token-level text.

To support legal language, the project uses a pre-trained legal language model and fine-tunes it for entity recognition on Dutch legal text.

Document workflow

A major part of the project was building a practical workflow that works for real users.

The final process:

1. A user uploads a Word (.docx) ruling
2. The system automatically labels entities and highlights them in the document
3. The labeled document is exported as a PDF for easy reading and sharing

This workflow supports a clean and consistent input/output format and allows the results to be reviewed in a familiar document layout.

What I delivered

• End-to-end NLP labeling workflow for Dutch legal rulings
• A document-based system that processes DOCX input and produces a labeled PDF
• Research and documentation to support the project decisions and improvements
• Final project report (Dutch)

Tools & Technologies

• *Python
• Transformers (Hugging Face)
• PyTorch
• Datasets
• Scikit-learn
• Seqeval
• Accelerate
• Safetensors