RAG Document Converter

📄 Convert documents to clean Markdown optimized for RAG pipelines

What This Actor Does

• Multi-format support - PDF, DOCX, PPTX, XLSX, HTML, images
• Structure preservation - Keeps headers, tables, lists intact
• RAG-optimized - Clean Markdown for LLM ingestion
• Section chunking - Split by headers for vector stores
• Metadata extraction - Title, author, page count

Use Cases

Use Case	Description
RAG Pipelines	Convert docs for retrieval-augmented generation
Knowledge Bases	Build searchable documentation
Content Migration	Convert legacy documents
LLM Context	Prepare documents for LLM analysis
Document Search	Index documents for semantic search

Input Examples

Basic PDF to Markdown

{
    "fileUrls": ["https://example.com/document.pdf"],
    "outputFormat": "markdown"
}

With Section Chunking

{
    "fileUrls": ["https://example.com/report.pdf"],
    "outputFormat": "markdown",
    "chunkBySection": true
}

Multiple Formats

{
    "fileUrls": [
        "https://example.com/doc.pdf",
        "https://example.com/slides.pptx",
        "https://example.com/data.xlsx"
    ],
    "outputFormat": "markdown"
}

With OCR

{
    "fileUrls": ["https://example.com/scanned.pdf"],
    "outputFormat": "markdown",
    "enableOcr": true
}

Configuration

Parameter	Type	Default	Description
fileUrls	array	-	Document URLs (required)
outputFormat	string	"markdown"	Output format
enableOcr	boolean	false	Use OCR for scanned docs
preserveTables	boolean	true	Convert tables
extractImages	boolean	false	Extract embedded images
chunkBySection	boolean	false	Split by headers
includeMetadata	boolean	true	Include doc metadata

Supported Formats

Format	Extensions
PDF	.pdf
Word	.docx
PowerPoint	.pptx
Excel	.xlsx
HTML	.html, .htm
Images	.png, .jpg, .jpeg, .tiff, .bmp

Output Formats

Format	Description
markdown	Clean Markdown (default, RAG-optimized)
html	HTML with structure
json	Lossless structured JSON
text	Plain text

Output

{
    "source": "https://example.com/document.pdf",
    "outputFormat": "markdown",
    "outputUrl": "https://api.apify.com/v2/key-value-stores/.../records/converted-12345.md",
    "contentPreview": "# Document Title\n\n## Introduction\n\nThis document covers...",
    "metadata": {
        "title": "Annual Report 2024",
        "pageCount": 42
    },
    "pageCount": 42,
    "success": true
}

With Section Chunking

{
    "source": "https://example.com/document.pdf",
    "sections": [
        { "title": "Introduction", "content": "..." },
        { "title": "Methodology", "content": "..." },
        { "title": "Results", "content": "..." }
    ],
    "sectionCount": 3,
    "success": true
}

RAG Integration

LangChain

from langchain.text_splitter import MarkdownTextSplitter

# Get markdown from actor output
markdown = result["contentPreview"]  # or fetch from outputUrl

splitter = MarkdownTextSplitter(chunk_size=1000)
chunks = splitter.split_text(markdown)

LlamaIndex

from llama_index import Document

doc = Document(text=markdown, metadata=result["metadata"])

Cost Estimation

Scale	Documents	Compute Units
Small	10	~0.05
Medium	50	~0.2
Large	200	~0.8

Technical Details

• Language: Python 3.12
• Library: IBM Docling
• Memory: 1GB-4GB (depends on document size)
• Features: 10x faster with DoclingParseV2

Limitations

• OCR requires additional processing time
• Very large documents may need more memory
• Some complex layouts may lose formatting

---

Keywords: docling, rag, pdf, markdown, convert, document, llm, retrieval, langchain, llamaindex