llm-agent-pdf-reader

A Python-based LLM Agent for Question Answering on PDF Documents

This project implements a Local Large Language Model (LLM) agent capable of answering questions based on the content of a provided PDF document. It leverages various LLM providers (like Gemini and OpenAI) and embedding techniques to understand and respond to user queries.

Features

• PDF Document Processing: Loads and chunks text content from PDF files.
• Multiple LLM Support: Integrates with both Google's Gemini and OpenAI models.
• Flexible Embedding Options: Supports Sentence Transformers and OpenAI embedding models.
• Vector Store for Retrieval: Utilizes FAISS for efficient similarity search of document chunks.
• Reranking for Relevance: Employs a CrossEncoder model to refine the selection of relevant context.
• Command-Line Interface (CLI): Provides a user-friendly CLI for executing the application with different configurations.
• JSON Output: Presents the question-answering results in a structured JSON format.

File Structure

llm-agent-pdf-reader/
├── .gitignore
├── README.md
├── run_agent.py        # CLI script with configurable LLM and embedding
├── requirements.txt    # Project dependencies
└── src/
    ├── __init__.py
    ├── config.py       # Configuration settings
    ├── document_processing/
    │   ├── __init__.py
    │   └── pdf_processor.py  # Handles PDF loading and chunking
    ├── embedding/
    │   ├── __init__.py
    │   └── embedding_models.py # Implements embedding models
    ├── llm/
    │   ├── __init__.py
    │   └── llm_agents.py     # Interfaces with LLM providers
    ├── question_answering/
    │   ├── __init__.py
    │   └── qa_agent.py       # Orchestrates question answering
    ├── retrieval/
    │   ├── __init__.py
    │   └── vector_store.py   # Implements the vector store
    └── utils/
        ├── __init__.py
        └── logger.py         # Sets up logging

Installation

1. Clone the repository:

   git clone <repository_url>
   cd llm-agent-pdf-reader

2. Create and activate a virtual environment (recommended):

   python -m venv venv
   source venv/bin/activate  # On Linux/macOS
   venv\Scripts\activate  # On Windows

3. Install dependencies:

   pip install -r requirements.txt

Configuration

1. Environment Variables: The application relies on environment variables for API keys and other settings. You can create a .env file in the project's root directory and populate it with the necessary values. Example .env file:

   GEMINI_API_KEY="YOUR_GEMINI_API_KEY"
   OPENAI_API_KEY="YOUR_OPENAI_API_KEY"
   EMBEDDING_MODEL_NAME="sentence-transformers/all-MiniLM-L6-v2"
   OPENAI_EMBEDDING_MODEL_NAME="text-embedding-3-large"
   LLM_MODEL_NAME="gemini-1.5-flash"
   OPENAI_LLM_MODEL_NAME="gpt-4o-mini"
   RERANKER_MODEL_NAME="cross-encoder/ms-marco-MiniLM-L-12-v2"
   PDF_PATH="documents/handbook.pdf"

   Note:

   • Obtain your API keys from Google AI Studio for Gemini and OpenAI Platform for OpenAI.
   • You can customize the model names and default PDF path in the .env file.

Usage

The primary way to run the application is through the run_agent.py script, which provides a flexible command-line interface.

python run_agent.py --help

This will display the available command-line options:

usage: run_agent.py [-h] [--pdf_path PDF_PATH]
                    [--llm_model {gemini,openai}]
                    [--embedding_model {sentence-transformers,openai}]
                    [--questions QUESTIONS] [--questions_file QUESTIONS_FILE]

LLM Agent PDF Reader CLI

options:
  -h, --help            show this help message and exit
  --pdf_path PDF_PATH   Path to the PDF file.
  --llm_model {gemini,openai}
                        Choose the LLM model (gemini or openai).
  --embedding_model {sentence-transformers,openai}
                        Choose the embedding model (sentence-transformers or
                        openai).
  --questions QUESTIONS
                        Semicolon-separated string of questions.
  --questions_file QUESTIONS_FILE
                        Path to a file containing questions (one question per
                        line).

Examples:

• Run with default settings (uses Gemini and default questions):

  python run_agent.py

• Specify a PDF file:

  python run_agent.py --pdf_path path/to/your/document.pdf

• Use OpenAI as the LLM:

  python run_agent.py --llm_model openai

• Use OpenAI for embeddings:

  python run_agent.py --embedding_model openai

• Provide questions directly:

  python run_agent.py --questions "What are the key benefits?, Explain the company culture."

• Read questions from a file:

  python run_agent.py --questions_file my_questions.txt

• Combine options:

  python run_agent.py --pdf_path another_document.pdf --llm_model openai --embedding_model openai --questions "What is the remote work policy?"