LlamaIndex Tutorial

This tutorial demonstrates how to build a Retrieval-Augmented Generation (RAG) system using LlamaIndex and open-source models.

The goal is to understand how LlamaIndex structures the RAG pipeline, including:

• document ingestion
• chunking
• embedding generation
• vector indexing
• retrieval
• LLM-based answer generation

The tutorial is divided into two parts:

1. API Tutorial – demonstrates the core LlamaIndex abstractions using a small toy dataset.
2. Example Tutorial – builds a complete RAG system using real-world text data.

Running the Tutorial with Docker¶

This tutorial can be executed inside Docker to ensure a fully reproducible environment.

Step 1 — Navigate to the Tutorial Directory¶

Open a terminal and run:

cd tutorials/tutorial_llamaindex

Step 2 — Build the Docker Image¶

Run:

docker build -t llamaindex-tutorial .

This command builds a Docker image containing:

• Python
• all project dependencies
• Jupyter Notebook

Step 3 — Run the Docker Container¶

Start the container with:

docker run -p 8888:8888 llamaindex-tutorial

After running the command, the terminal will display a link similar to:

http://127.0.0.1:8888/?token=...

Open this link in your browser to access the Jupyter notebook environment.

Step 4 — Run the Notebooks¶

Inside Jupyter, run the notebooks in the following order:

1. llamaindex.API.ipynb
2. llamaindex.example.ipynb

Tutorial Structure¶

tutorial_llamaindex/
│
├── llamaindex.API.ipynb
├── llamaindex.example.ipynb
├── llamaindex.API.md
├── llamaindex.example.md
├── llamaindex_utils.py
├── Dockerfile
├── requirements.txt
└── data/

API Tutorial¶

File:

llamaindex.API.ipynb

This notebook demonstrates the core components of LlamaIndex using a small toy dataset.

Topics covered:

• loading documents with SimpleDirectoryReader
• splitting documents into nodes
• generating embeddings
• building a VectorStoreIndex
• querying the index
• inspecting retrieved nodes
• customizing prompts

The objective of this notebook is to understand how LlamaIndex APIs work independently.

Example Tutorial¶

File:

llamaindex.example.ipynb

This notebook demonstrates a complete RAG pipeline using real-world text data.

The notebook automatically downloads public domain books from Project Gutenberg, including:

• Pride and Prejudice
• The Adventures of Sherlock Holmes

The workflow includes:

1. downloading text data
2. loading documents
3. splitting documents into chunks
4. generating embeddings
5. building a vector index
6. retrieving relevant passages
7. generating answers using an LLM

This notebook shows how LlamaIndex can be used to build a real-world RAG system.

Models Used¶

The tutorial uses open-source models so it can run locally without requiring API keys.

Embedding model:

sentence-transformers/all-MiniLM-L6-v2

Language model:

microsoft/phi-2

Notes¶

Large datasets are not stored in the repository.

Instead, the notebooks automatically download the required texts when executed for the first time.
This keeps the repository lightweight while ensuring reproducibility.

Summary¶

This tutorial demonstrates how LlamaIndex simplifies the RAG architecture by providing abstractions for:

• document ingestion
• chunking
• vector indexing
• retrieval
• prompt orchestration

Compared to a naive implementation, LlamaIndex reduces boilerplate code and provides a modular framework for building LLM-powered applications.