RAG Integration

Why EdgeParse for RAG?

EdgeParse produces structured output with reading order, heading hierarchy, and table structure — all critical for high-quality retrieval-augmented generation.

LangChain Integration

import edgeparse
from langchain.text_splitter import MarkdownHeaderTextSplitter

# Extract structured Markdown
markdown = edgeparse.convert("document.pdf", format="markdown")

# Split by heading hierarchy
splitter = MarkdownHeaderTextSplitter(
    headers_to_split_on=[
        ("#", "H1"),
        ("##", "H2"),
        ("###", "H3"),
    ]
)

chunks = splitter.split_text(markdown)

for chunk in chunks:
    print(chunk.metadata)
    print(chunk.page_content[:100])
    print("---")

LlamaIndex Integration

import edgeparse
from llama_index.core import Document

# Extract as JSON for maximum metadata
json_str = edgeparse.convert("document.pdf", format="json")
import json
data = json.loads(json_str)

# Create documents with rich metadata
documents = []
for element in data["kids"]:
    if element["type"] in ("paragraph", "heading"):
        doc = Document(
            text=element["content"],
            metadata={
                "page": element["page number"],
                "type": element["type"],
                "source": data["file name"],
            }
        )
        documents.append(doc)

Best Practices

1. Use Markdown format for heading-based chunking
2. Use JSON format when you need bounding boxes or element metadata
3. Enable safety filters to prevent prompt injection from malicious PDFs
4. Preserve table structure — structured tables improve retrieval quality