parent-child-chunker 0.1.0

Deterministic parent–child chunking for Markdown documents

Parent Child Chunker

Deterministic parent–child chunking for Markdown documents, designed for Retrieval-Augmented Generation (RAG) pipelines.

This library takes structured Markdown as input and produces:

• parent chunks aligned with document hierarchy
• child chunks optimized for vector search
• explicit, traceable parent–child relationships

The core is lightweight, dependency-free, and framework-agnostic.

---

Why parent–child chunking?

Naive text splitting often breaks semantic structure and loses context. Parent–child chunking preserves document hierarchy while enabling fine-grained retrieval.

Typical use cases:

• RAG pipelines
• documentation indexing
• knowledge base ingestion
• long-form Markdown processing

---

What this library does

• Accepts raw Markdown text
• Splits content by header hierarchy
• Normalizes parent chunks by size
• Generates child chunks with stable parent references
• Produces deterministic chunk identifiers

---

What this library does not do

• PDF or document ingestion
• OCR
• layout reconstruction
• semantic embedding or ranking

Those concerns are intentionally left to upstream tools.

---

Installation

pip install parent-child-chunker

# OPTIONAL LangChain adapter:
pip install parent-child-chunker[langchain]

---

Basic usage from parent_child_chunker import ParentChildMarkdownChunker

markdown_text = """

Introduction

This is an introduction.

Background

Some background information.

Details

More detailed content. """

chunker = ParentChildMarkdownChunker( min_parent_chars=500, max_parent_chars=3000, child_chunk_size=512, child_overlap=64, )

parents, children = chunker.chunk(markdown_text)

print(f"Parents: {len(parents)}") print(f"Children: {len(children)}")

---