🇪🇬 arabic-pdf-rag

Arabic-First PDF Extraction, RTL Normalization & Table Chunking Engine for RAG Pipelines.

📌 The Problem

Standard PDF readers (PyPDF, PDFMiner, Unstructured) mangle Arabic text:

• ❌ Arabic letters come out disconnected (م ح م د instead of محمد).
• ❌ Sentences are reversed from Left-to-Right instead of Right-to-Left (RTL).
• ❌ Digits and mixed Arabic/English technical terms get garbled.
• ❌ Vector embeddings in Qdrant/Chroma store corrupted text, leading to failed RAG retrieval.

arabic-pdf-rag solves this completely in 2 lines of Python code!

⚡ Quick Start

pip install arabic-pdf-rag

Usage Example:

from arabic_pdf_rag import ArabicRAGExtractor

extractor = ArabicRAGExtractor()

# Extract and normalize Arabic PDF for RAG
documents = extractor.extract_pdf("egyptian_law_document.pdf")

for doc in documents:
    print(f"Page {doc.page_number}:")
    print(doc.clean_text)
    
# Generate RAG-ready chunks
chunks = extractor.chunk_document(documents, chunk_size=500, overlap=50)
print(f"Generated {len(chunks)} RAG chunks ready for Vector DB!")

🛠️ Features

• 🔄 Automatic RTL Reshaping & Bidi Normalization.
• 📄 Multi-Format Extraction: PDF, scanned text, and tables.
• 📑 Smart Paragraph & Semantic Chunking tailored for Arabic syntax.
• 🔒 100% Offline & Air-Gapped Safe: Zero external API calls.

👤 Author

Mohamed Shaban (محمد شعبان العتماني) — Applied AI Engineer
🌐 Website: msalatmani.org | 📧 Email: msalatmani@gmail.com