DocBlocks — Layout-Preserving PDF Text Extraction

DocBlocks is a Python library for extracting text from digitally generated PDFs while preserving their visual reading order and page structure. It avoids the alignment distortions found in traditional PDF extractors.

Note: DocBlocks works only with computer-generated (digital) PDFs.
It does not support scanned/image PDFs at this time.

---

Purpose

Most extraction libraries produce output that is:

• misaligned
• reordered
• merged across columns
• missing sections

DocBlocks solves this by providing:

• accurate reading order
• non-destructive extraction
• page-based segmentation
• zero content modification

Final mission: no more ugly alignment at the end.

---

Guarantees

DocBlocks guarantees:

• no text modification
• no text deletion
• no page merging
• no reordering beyond true reading order
• one page’s content will never interfere with another page

This ensures perfect segmentation for AI/ML pipelines, RAG systems, or document processing.

---

Current Limitations

DocBlocks does not support:

• scanned/image PDFs
• OCR-based text extraction
• non-PDF formats (Word, images, HTML)

This limitation is intentional to guarantee perfect handling of digital text PDFs.

---

Roadmap (Planned for End of February Release)

The upcoming version will include:

• support for additional document formats
• OCR engine for scanned documents
• richer layout information
• improved block detection

OCR engine names are intentionally not mentioned — expect powerful, production-grade OCR.

---

Installation

pip install docblocks

Requires Python 3.8+

---

Usage

Extract text into memory

import docblocks

text = docblocks.extract(r"document.pdf")
print(text)

Convert PDF → TXT file

import docblocks

docblocks.convert(
    r"document.pdf",
    r"output.txt"
)

You must use r"..." (raw strings) for Windows paths to avoid escape issues.

---

Output Format

DocBlocks ensures:

• clean page segmentation
• pure reading order
• no cross-page interference
• no ugly alignment errors

Example output:

┌──────── PAGE 1 ─────────┐
This is page one content...
└──────────────────────────┘

┌──────── PAGE 2 ─────────┐
This is page two content...
└──────────────────────────┘

Notice that:

Page 1 content never flows into Page 2, and Page 2 never leaks into Page 1.

---

Best For

DocBlocks is ideal for:

• research papers
• legal & policy docs
• technical manuals
• structured reports
• AI preprocessing (RAG / LLM input)

---

License

MIT — free for commercial & academic use.

---

Vision

DocBlocks exists for one purpose:

“No more ugly alignment at the end.”

---

Contributing

Issues and suggestions are welcome:

https://github.com/vishalp-dev24/DocBlocks.git

Pull requests, feedback, and feature requests are appreciated.