Vectorless RAG: How I Built a RAG System Without Embeddings, Databases, or Vector Similarity

Vectorless RAG Implementation: Document Tree Generation

The goal is to convert a PDF document into a structured hierarchical tree representation that can be navigated and reasoned over. We'll use the pymupdf4llm library for layout-aware parsing, which helps in preserving the structure of headings and content.

Step 1: Extract Structured Markdown

First, we extract markdown text from the PDF while preserving its layout and heading levels using pymupdf4llm.

python
 
1import fitz  # PyMuPDF
2from pymupdf4llm import to_markdown
3
4def extract_markdown(pdf_path):
5    doc = fitz.open(pdf_path)
6    markdown_content = ""
7    
8    for page_num in range(doc.page_count):
9        text = doc.load_page(page_num).get_text("text")
10        markdown_content += to_markdown(text) + "\n"
11        
12    return markdown_content

Step 2: Parse Markdown Headers into Tree Hierarchy

Next, we parse the extracted markdown content to build a hierarchical tree structure.

python
 
1import re
2
3def parse_markdown_to_tree(markdown_content):
4    lines = markdown_content.split("\n")
5    stack = []
6    root
7
8[Read the full article at Towards AI - Medium](https://pub.towardsai.net/vectorless-rag-how-i-built-a-rag-system-without-embeddings-databases-or-vector-similarity-efccf21e42ff?source=rss----98111c9905da---4)
9
10---
11
12**Want to create content about this topic?** [Use Nemati AI tools](https://nemati.ai) to generate articles, social posts, and more.