Turn preprints into simulation datasets automatically using LLMs

[Essmaw]

Context

Currently, manually writing titles and descriptions for simulation datasets is time-consuming and prone to errors. We need a tool that leverages LLMs to extract metadata directly from a preprint PDF to automate the creation of repository-ready records.

Objective

Create a pipeline that takes a preprint as input and outputs:

1. A Repository-Ready Title & Description: Specifically designed for data repositories (like Zenodo or Materials Cloud), ensuring the dataset is discoverable and well-documented.
2. A Structured Config File: A formatted file containing the technical keys required for reproduction.