Resume Parser Dataset

This project prepares a resume-parsing dataset through four stages:

• .data/raw/ — source resume files, organized by job category.
• .data/extracted/ — text extracted from the source files.
• .data/annotated/ — schema-validated target JSON for resumes.
• .data/curated/ — train, validation, and test JSONL splits.

Local setup

From the project root, create and activate a virtual environment, then install the dependencies:

python3 -m venv .venv
source .venv/bin/activate
python -m pip install --upgrade pip
pip install -r requirements.txt

Copy .env.example to .env and set HF_DATASET_REPO to the Hugging Face dataset ID, for example md-zain/resume-parser-dataset. A Hugging Face token is optional for downloading a public dataset; it is required to upload changes.

Download or update the dataset locally with:

python download_resume_parser_dataset.py

The script downloads the configured dataset into .data/ and displays a progress bar. It updates files from the remote repository without deleting local-only files.

To upload the complete .data/ pipeline to the configured public Hugging Face dataset repository, run:

python upload_resume_parser_dataset.py

The uploader reports the file count and size, then uses Hugging Face's Xet transfer progress display. It requires a valid HF_TOKEN with write access.

Dataset sources

• Mehyaar – Annotated NER PDF Resumes
• opensporks – Resumes
• hadikp – Resume Dataset PDF