Simple Vector Index Example

Building a Custom RAG Pipeline with LlamaIndex and Llama 3.2

This example shows how to build a classic Retrieval-Augmented Generation (RAG) pipeline with LlamaIndex: parse a PDF document, embed it with a Hugging Face model, index the embeddings in a vector store, and answer questions with Groq's Llama 3.2 LLM. The vector store is persisted to disk so it can be reused without re-indexing.

What the notebook covers

• Loading and parsing a PDF document with LlamaParse
• Creating vector embeddings with a Hugging Face model (sentence-transformers/all-MiniLM-L6-v2)
• Building a VectorStoreIndex and querying it with Groq's Llama 3.2 LLM
• Persisting the vector store to ./storage/ for reuse

Prerequisites

• Python 3.11+
• A Groq API key
• A LlamaIndex Cloud (LlamaParse) API key
• A Hugging Face token

Setup

1. Create and activate a Python virtual environment:

   python -m venv .venv
   source .venv/bin/activate      # On Windows: .venv\Scripts\activate

2. Install the required packages:

   pip install -r requirements.txt

3. Create your environment file from the template:

   cp .env.example .env

4. Edit .env and add your API keys:

   • GROQ_API_KEY — your Groq API key
   • LLAMA_CLOUD_API_KEY — your LlamaIndex Cloud (LlamaParse) API key
   • HF_TOKEN — your Hugging Face token

Running the demo

1. Launch Jupyter and open main.ipynb:

   jupyter notebook main.ipynb

2. Run all cells to see the demo in action.

Notes

• Make sure all required API keys are set in .env before running the notebook.
• The demo ships with a sample PDF (data/2025_Tucson_Hybrid_user_manual.pdf). Point the loader at your own document to run the pipeline over your own data.

---

Written and developed by Amin Amiri. Released under the MIT License.