legalis

This is the mono repo for the legalis project. The focus of the project is prediction work and data science around German Court cases. Currently, this repo contains a folder for Data, Model, and App. Each contains the notebook/python files for the respective part of the project.

The project contains two notebooks for cleaning and enhancing with the CHatGPT API, as well as one notebook used for training a random forest and Naive Bayes Multimodal Model and one for a BERT Classification Model fine-tuned on top of a German BERT. The latter is also running in hugging face spaces.

ℹ️ This is an archive of tasks done as part of my studies.

Ultimately I was able to predict the outcome of 60% of the court cases. But to be fair, the methods are not completely scientifically corrent.

🔗 Links:

Interactive Model Demos on Huggingface

BERT Model on Huggingface

Random Forest Classifier on Huggingface

Labeled Dataset on Huggingface

🏗️ Tech Stack:

Language/Platform: Python, JupyterNotebook

Noteable Libaries: 🤗Datasets, 🤗Evaluate, scikit-learn, 🤗Transformers

Base Dataset: OpenLegalData

👨‍💻 Author and Credits:

Author: @LennardZuendorf

• This project is part of a course in HTW Berlin's Business Computing Bachelor Program ("Unternehmenssoftware")