LlamaIndex Document Query System

This project demonstrates a simple document querying system using LlamaIndex, a powerful tool for building AI-powered applications with large language models.

What Will We Build?

• We will build a simple document querying system that allows you to ask questions about a set of documents and get answers.
• We will use a language model to answer questions.
• We will use a vector store index to store and query the documents.
• Create a RAG application (Query Engine)
• Create a simple chatbot (ChatEngine)
• Create the engine code first, then the chatbot loop to talk to the engine

  while True:
    text_input = input("User: ")
    if text_input == "exit":
        break
    response = agent.chat(text_input)
    print(f"Agent: {response}")

• Observability (Logging, Metrics, Traces) - https://docs.llamaindex.ai/en/stable/module_guides/observability/
• Documents vs. Nodes
• LlamaParse
• LlamaHub
• LlamaCloud (is a waitlist, but promising features)
• NEEDED:
  • OPENAI_API_KEY
  • PHOENIX_API_KEY (https://llamatrace.com/login)

Overview

The main script (llamaindex/main.py) performs the following tasks:

1. Loads documents from a PDF directory
2. Creates or loads a vector store index
3. Executes a query on the indexed documents

Dependencies

• LlamaIndex
• OpenAI API (for the underlying language model)

Code Explanation

Environment Setup

The project uses pip to manage dependencies. To install the required packages, run:

pip install llama-index openai

Loading Documents

The load_documents function in main.py uses SimpleDirectoryReader from LlamaIndex to load all PDF files from the specified directory.

Creating or Loading a Vector Store Index

The create_index function in main.py creates a new vector store index if it doesn't exist, otherwise, it loads the existing index.

Query Execution

The query_index function in main.py uses the OpenAI API to execute a query on the indexed documents.

Usage

To run the project, execute the following command:

python main.py

This will load the documents, create or load the index, and execute the query.