Vectorless RAG Build a Complete RAG System from Scratch

BM25 + PageIndex + Ollama - A Complete RAG System
What you'll learn

---

You can build a Complete vectorless RAG System Without Vectors or Embeddings
You will implement BM25 - to find the most relevant documents across a folder of PDFs, learn about the k1 and b parameters and how to tune them.
You will extract headings from documents and build a hierarchical tree structure and implement tree navigation - step by step - using LLM reasoning
You will combine BM25 and PageIndex into a complete hybrid pipeline and You will write production-ready Python modules - not just Jupyter notebooks.
Requirements

---

You should be comfortable with Python fundamentals.
You should be comfortable using the terminal or command prompt.
It helps if you know what RAG is - but it's not required.
You'll be running a local LLM using Ollama, 8GB of RAM is the minimum recommendation. 16GB is better
Description

---

## Course Description

---

- 200 Words
---
Build a complete RAG system without vectors, embeddings, or vector databases.
In this course, you'll build BM25 for lexical search, PageIndex for document navigation, and a complete retrieval pipeline that works on real PDFs.
No API keys. No cloud costs. No expensive vector infrastructure. So you can experiment freely without worrying about the API cost.
You'll understand what is happening underneath and will be able to debug or improve the retrieval engine - because you build the retrieval components yourself.
You'll build them. You'll tune them. You'll benchmark them.
And you'll learn how to determine why one retrieval strategy performs better than another. You'll learn how to handle real-world document retrieval, tune BM25 parameters, navigate large structured documents, and improve retrieval when a basic search approach isn't giving you satisfactory results.
You'll also learn how different retrieval techniques behave on different types of documents and queries, helping you make better engineering decisions instead of simply choosing a retrieval method by default.
By the end, you won't just know how to build a RAG application. You'll understand how to engineer the retrieval system behind it.
Everything runs locally with Python, BM25, PageIndex, and Ollama. No cloud. No GPU. No hidden costs.
Just you, your code, and a complete understanding of how retrieval actually works.
Who this course is for

---

Students and Job Seekers Building Portfolio Projects
Professionals Working with Structured Documents
You're building an AI product but don't have a big budget. You can't afford expensive API calls or cloud services.
Homepage

---