LLM Task

Gemini PDF Question Answering System

This CLI app let's users upload, process it, and ask questions on the content using a conversational chat powered by Google's Gemini-pro.

Prerequisites

• Python 3.10 or higher
• PyPDF2
• Langchain
• Google Generative AI
• Google Cloud SDK

Installation

1. Clone this repository to your local machine.
2. Create a virtual environment and activate it.
3. Install the required packages by running pip install langchain google-generativeai faiss-cpu python-dotenv PyPDF2 langchain_google_genai
4. Create a .envfile in the root directory of the project and add your Google API key:

GOOGLE_API_KEY=your_google_api_key

5. Run the application by executing python new_app [file_name]
   • change the file_name with the pdf name that you want to work on.
6. After processing the PDF, the application will enter an interactive mode where you can ask questions.
   • the answers will come in command line like this
   • there will be on-screen commands
   • if the question does not have any answer from the provided pdf then the response will be blank.
   • type 'exit' to quit the program

Code Structure of the main file

1. get_pdf_text: Extract Text from PDF Files

• Purpose: This function extracts text from the uploaded PDF files.
• How It Works: It takes a list of PDF file paths, reads each file using PyPDF2, and concatenates all the text from each page. The resulting string contains the complete content from all PDFs combined.

2. get_text_chunks: Split Extracted Text into Chunks

• Purpose: This function splits the extracted text into smaller, manageable chunks.
• How It Works: Using the RecursiveCharacterTextSplitter from Langchain, the function divides the text into chunks of 10,000 characters each, with a 1,000-character overlap between consecutive chunks. This ensures that context is preserved across chunks for better question answering.

3. get_vector_store: Create a Vector Database for Text Chunks

• Purpose: This function creates a vector database for the text chunks.
• How It Works: It uses the embeddings from Google Generative AI to transform each text chunk into a vector representation. These vectors are then stored using FAISS, which is a library for efficient similarity search. The vector database is saved locally as "faisss_index".

4. get_conversational_chain: Create a Conversational Chain

• Purpose: This function creates a conversational chain to be used for answering questions.
• How It Works: It defines a prompt template for how to respond to user queries. It uses the ChatGoogleGenerativeAI model (gemini-pro) and creates a question-answering chain (load_qa_chain) that combines context and user input to generate a detailed response.

5. process_user_input: Handle User Input and Generate Answers

• Purpose: This function processes user input, searches the vector store, and generates a response.
• How It Works: When the user provides a question, this function retrieves relevant chunks from the FAISS vector store by performing a similarity search. It then uses the conversational chain to generate an answer based on the user question and the most relevant parts of the extracted PDF text.

6. main: Set Up the Command-Line Application and Run the User Interaction Loop

• Purpose: This is the main function that sets up the CLI program and runs the user input process.
• How It Works: It uses argparse to parse command-line arguments (the list of PDF file paths). After extracting text, splitting it into chunks, and storing the vectors, it enters a loop where the user can ask questions about the content of the processed PDF files. The loop will continue until the user types 'exit'.