README.md

metadata

license: apache-2.0
language:
  - fr
  - en
task_categories:
  - visual-document-retrieval
  - text-retrieval
tags:
  - RAG
  - DSE
  - retrieval
  - geotechnie
configs:
  - config_name: train
    data_files: train-*.parquet
  - config_name: filtered
    data_files: filtered-*.parquet

OGC - Organized, Grouped, Cleaned

Geotechnie Vision DSE

Intended for image/text to vector (DSE)

Dataset Composition

Made with https://github.com/RacineAIOS/OGC_pdf-to-parquet

This dataset was created by scraping PDF documents from online sources and generating relevant synthetic queries. We used Google's Gemini 2.0 Flash Lite model in our custom pipeline to produce the queries, allowing us to create a diverse set of questions based on the document content.

Dataset Statistics

Metric	Value
Total rows	≈ 69k

Language Distribution

Language	Ratio (%)
French (fr)	90
English (en)	10

Creators

Dataset curated by:

• Paul Lemaistre
• Léo Appourchaux