ismailcemsahin/job-titles-descriptions

Synthetic Job Descriptions Dataset

A high-quality synthetic expansion of gpriday/job-titles — 65,248 structured job descriptions with a pre-computed FAISS index for semantic search.

Overview

This dataset provides comprehensive, synthetically generated job descriptions for all role titles found in the gpriday/job-titles source dataset. Each entry is structured for consistency and designed to support NLP pipelines, career-tech applications, and Retrieval-Augmented Generation (RAG) systems.

Generation Methodology

Job titles were sourced from the gpriday/job-titles repository. Llama 3.1-8B was used to synthesize professional, contextually grounded descriptions for all 65,248 roles. Each description follows a consistent structure to ensure terminological clarity and relevance to its corresponding title.

File Structure

.
├── train.parquet        # Primary dataset file
└── job_index.faiss      # Pre-computed FAISS index (dim=768)

train.parquet

job_index.faiss

Pre-computed FAISS index optimised for semantic similarity search and vector-based retrieval. Index dimension: 768.

Usage

Loading the Dataset

from datasets import load_dataset

dataset = load_dataset("ismailcemsahin/job-titles-descriptions")

# Inspect the first record
print(dataset['train'][0])

Semantic Search with FAISS

# Attach the pre-computed FAISS index
dataset.load_faiss_index('embeddings_column_name', 'job_index.faiss')

# Run a nearest-neighbour search
# scores, samples = dataset.get_nearest_examples(
#     'embeddings_column_name', query_vector, k=5
# )

Note: Replace 'embeddings_column_name' with the name of your embeddings column and query_vector with a numpy array of shape (768,).

Use Cases

• —Semantic job search — find roles by meaning, not just keywords
• —Career platform recommendations — match candidates to relevant titles
• —RAG pipelines — ground LLM responses with structured role context
• —NLP research — training and evaluation of job-domain models

Citation & Acknowledgements

This dataset is a synthetic expansion built on top of community contributions. Credit to gpriday for releasing the foundational job-titles dataset to the open-source community.

@dataset{sahin2026jobdescriptions,
  author    = {Ismail Cem Sahin and Greg Priday},
  title     = {Synthetic Job Descriptions Dataset (65k Roles)},
  year      = {2026},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasets/ismailcemsahin/job-titles-descriptions}
}

Author: Ismail Cem Sahin License: Apache 2.0