RAG Chunking Strategy Simulator

An interactive Enterprise AI tool to learn, compare, visualize, and evaluate chunking strategies for Retrieval-Augmented Generation (RAG).

---

📖 Overview

Chunking is one of the most important—but often overlooked—steps in a RAG pipeline.

Different document types require different chunking strategies to maximize retrieval quality and LLM response accuracy.

This simulator allows developers, AI engineers, and architects to:

• Compare multiple chunking strategies
• Visualize chunk boundaries
• Evaluate chunk quality
• Explore metadata generated during chunking
• Understand when to use each strategy
• Generate LangChain code snippets
• Benchmark chunking approaches for enterprise AI systems

---

✨ Features

📚 Supported Chunking Strategies

Basic

• Fixed Character
• Fixed Word
• Fixed Token
• Sliding Window
• Paragraph
• Sentence

Adaptive Hybrid

LangChain

• Recursive Character Splitter
• Markdown Header Splitter
• HTML Header Splitter
• Recursive JSON Splitter
• Python Code Splitter
• JavaScript Code Splitter

Overlapping Chunking

Hierarchical Chunking

Summary Attached Chunking

Metadata Enhanced

Parent Child

AI-powered

• Semantic Similarity Chunking
  • Sentence Transformers
  • Configurable similarity threshold
• LLM based chunking

---

📄 Supported File Types

✅ TXT

✅ Markdown

✅ PDF

✅ DOCX

✅ HTML

✅ JSON

✅ CSV

✅ Python

✅ JavaScript

✅ TypeScript

✅ JSX / TSX

---

📊 Analysis Dashboard

After processing a document the simulator displays

• Total Chunks
• Average Characters
• Minimum Chunk Size
• Maximum Chunk Size
• Average Words

---

📈 Quality Evaluation

The simulator automatically evaluates every chunking strategy.

Chunk Quality Score

Evaluates

• Chunk size consistency
• Boundary quality
• Readability

---

Context Preservation Score

Measures how well context is preserved across chunk boundaries.

---

Metadata Score

Shows how much useful metadata has been generated.

---

🏷 Metadata Visualization

Each chunk can include metadata such as

• Chunk Type
• Break Reason
• Similarity Score
• Language
• Section Information

---

🤖 AI Recommendation Engine

The simulator analyzes the document and recommends the most suitable strategy.

Document	Recommendation
Plain Text	Recursive
Markdown	Markdown Header
HTML	HTML Header
JSON	Recursive JSON
Python	Python Code
JavaScript	JavaScript Code
Multi-topic Text	Semantic Similarity

---

⚖ Strategy Comparison

Compare every chunking strategy using

• Total Chunks
• Average Chunk Size
• Minimum Size
• Maximum Size
• Average Words

---

🔄 Recursive vs Semantic Comparison

Visual side-by-side comparison of

• Recursive Character Splitter
• Semantic Similarity Chunker

Compare

• Chunk boundaries
• Metadata
• Similarity
• Chunk sizes

---

🛠 Developer Tools

• Export JSON
• Copy Chunks
• Generate LangChain Code
• Copy LangChain Code

---

🧪 Testing

Unit tests cover

• Fixed Character Chunker
• Fixed Word Chunker
• Sliding Window Chunker
• Paragraph Chunker
• Sentence Chunker
• Recursive Chunker
• Markdown Chunker
• HTML Chunker
• JSON Chunker
• Python Chunker
• JavaScript Chunker
• Semantic Chunker
• Adaptive Chunker
• Metadata Chunker
• Parent Child Chunker
• Summary Attached Chunker
• LLM assisted Chunker

Run all tests

cd backend

source venv/bin/activate

pytest -v

---

🏗 Architecture

                  +---------------------+
                  |    Next.js UI       |
                  +----------+----------+
                             |
                             |
                  REST API (FastAPI)
                             |
      +----------------------+----------------------+
      |                                             |
      |                                             |
Chunk Service                              Upload Service
      |
      |
+---------------------------+
| Chunking Strategy Layer   |
+---------------------------+
| Fixed                     |
| Recursive                 |
| Sliding                   |
| Paragraph                 |
| Sentence                  |
| Markdown                  |
| HTML                      |
| JSON                      |
| Code                      |
| Semantic                  |
+---------------------------+
             |
             |
Evaluation Engine
             |
             |
Metadata + Statistics
             |
             |
JSON Response

---

📂 Project Structure

rag-chunking-simulator/

├── backend/
│   ├── app/
│   │   ├── api/
│   │   ├── chunkers/
│   │   ├── services/
│   │   ├── utils/
│   │   └── main.py
│   │
│   ├── tests/
│   ├── requirements.txt
│   └── pytest.ini
│
├── frontend/
│   ├── app/
│   ├── components/
│   ├── public/
│   └── types/
│
└── README.md

---

🚀 Installation

Backend

cd backend

python -m venv venv

source venv/bin/activate

pip install -r requirements.txt

uvicorn app.main:app --reload

Backend

http://localhost:8000

Swagger

http://localhost:8000/docs

---

Frontend

cd frontend

npm install

npm run dev

Frontend

http://localhost:3000

---

📸 Screenshots

Dashboard

Strategy comparison

AI recommendation

Chunking

Recursive vs Sementic comparison

Developer Tool

Metadata Filter

---

🗺 Roadmap

✅ Completed

• Multiple chunking strategies
• Semantic chunking
• File upload
• Strategy comparison
• Recursive vs Semantic comparison
• Recommendation engine
• Metadata visualization
• Chunk quality metrics
• Export JSON
• LangChain code generation
• Unit tests

🚧 Planned

• Retrieval Simulator
• Embedding Visualization
• RAPTOR Chunking
• OCR-aware Chunking
• Multi-document Benchmarking
• Docker
• GitHub Actions CI/CD
• Vector Database Integration
• RAG Evaluation Dashboard

---

🤝 Contributing

Contributions, ideas, and feature requests are welcome.

1. Fork the repository
2. Create a feature branch
3. Commit your changes
4. Open a Pull Request

---

📜 License

MIT License

---

👨‍💻 Author

Jay Ram Singh

AI Engineer | Enterprise AI Architect | RAG & LLM Systems

• GitHub: https://github.com/code-jay
• LinkedIn: https://www.linkedin.com/in/jayram/

---

⭐ If you find this project useful, consider giving it a Star!