Natural language processing course: Automatic generation of Slovenian traffic news for RTV Slovenija

This project aims to automate the generation of concise, accurate traffic news for RTV Slovenija. The current process relies on manual reporting every 30 minutes, which is both time-consuming and prone to error. By leveraging state-of-the-art large language models, prompt engineering techniques, and parameter-efficient fine-tuning methods, the project seeks to deliver real-time, high-quality traffic updates.

Project Folder Structure

.
├── arnes/      # Everything related to running LLMs on ARNES
├── Articles/   # Related works
├── Data/       # Data files
├── Notebooks/  # Relevant notebooks for our project
├── report/
└── src/        # Preprocessing & evaluation code