opendataloader

OpenDataLoader PDF — Tutorial Suite

Sample programs covering the major features of opendataloader-pdf (v2.4.1, Apache 2.0).

Prerequisites

Requirement	Notes
Java 11+ on PATH	conda install -n genai -c conda-forge openjdk=11
opendataloader-pdf	already in genai conda env
langchain-opendataloader-pdf	already in genai conda env (tutorial 10 only)

Activate the environment before running any tutorial:

conda activate genai

For a fresh environment instead:

conda env create -f environment.yml
conda activate opendataloader

Test data (data/)

File	Content
sample.pdf	Generic multi-page document
Systems Thinking in 25 Words or Less.pdf	Well-structured text document
Systems Thinking Four Key Questions.pdf	Multi-heading structured document
NVIDIAAn.pdf	Corporate document with images and URLs
table.pdf	Document with tabular data
scanned.pdf	Image-only scanned PDF (needs OCR)

Tutorials

#	File	Feature	Key parameter(s)
01	tutorial_01_basic_extraction.py	Single-file Markdown + text	format="markdown", format="text"
02	tutorial_02_output_formats.py	All four output formats	format="markdown,json,html,text"
03	tutorial_03_json_structure.py	JSON schema analysis — element types, headings, bounding boxes, fonts	format="json"
04	tutorial_04_table_extraction.py	Table element isolation	format="json,markdown" on table.pdf
05	tutorial_05_image_extraction.py	Image output modes	image_output="off/external/embedded", image_format="png/jpeg"
06	tutorial_06_batch_processing.py	Whole-folder vs file-list batch	input_path=[folder] vs input_path=[list]
07	tutorial_07_sanitization.py	PII scrubbing for AI safety	sanitize=True
08	tutorial_08_struct_tree.py	Native PDF tags + tagged-PDF export	use_struct_tree=True, format="tagged-pdf"
09	tutorial_09_scanned_ocr.py	OCR for scanned PDFs (hybrid mode)	hybrid="docling-fast"
10	tutorial_10_langchain_rag.py	LangChain RAG pipeline	OpenDataLoaderPDFLoader, FAISS

Running

# Run a single tutorial
conda activate genai
cd Code/opendataloader
python tutorial_01_basic_extraction.py

# Run all (non-OCR) tutorials
for i in 01 02 03 04 05 06 07 08 10; do
    python tutorial_0${i}_*.py
done

Each tutorial writes its output under output/<nn>_<name>/.

Tutorial 09 — OCR special setup

Tutorial 09 requires the hybrid backend running in a separate terminal:

conda activate genai
opendataloader-pdf-hybrid --port 5002 --force-ocr

Then in another terminal:

conda activate genai
cd Code/opendataloader
python tutorial_09_scanned_ocr.py

The script prints graceful instructions if the backend is not reachable.

Output formats reference

Format	File ext	Best for
markdown	.md	LLM context / RAG ingestion
json	.json	Structured access (bounding boxes, element types)
html	.html	Web display
text	.txt	Plain string processing
tagged-pdf	.pdf	Accessibility-compliant PDF/UA output
annotated-pdf	.pdf	Visual debugging of detected structure