Skip to content

README.md

Latest commit

 

History

History
116 lines (69 loc) · 4.44 KB

README.md

File metadata and controls

116 lines (69 loc) · 4.44 KB

Langchain RAG Chatbot

Project Description

The Langchain RAG Chatbot is an advanced AI-powered conversational tool designed to provide context-aware responses by leveraging Retrieval-Augmented Generation (RAG). This project combines document retrieval with language model generation, enabling users to upload documents in various formats (PDF, DOCX, HTML), index their contents, and interact with a chatbot to retrieve and discuss information from those documents. Built with a modern tech stack and a sleek Streamlit interface, this application is ideal for users who need an efficient way to query and explore document-based knowledge.

Tech Stack

The project utilizes a robust set of technologies, each serving a specific purpose:

Frontend:

Streamlit: A Python library for creating an interactive web-based user interface. It powers the chat window, sidebar, and document management features.

Backend:

FastAPI: A high-performance web framework for building APIs with Python. It handles requests for document processing, chat interactions, and database operations.

Database:

MongoDB: A NoSQL database used to store chat logs (conversation history) and metadata about uploaded documents, such as file names and indexing status.

Vector Store:

ChromaDB: A lightweight, embedded vector database for storing and retrieving document embeddings, enabling fast similarity searches for RAG.

Language Model:

Google Generative AI (Gemini-1.5-Flash): A powerful generative AI model used to produce human-like responses based on retrieved document chunks.

Embedding Model:

HuggingFace all-MiniLM-L6-v2: A sentence transformer model for generating embeddings from document text, ensuring efficient retrieval of relevant content.

Document Loaders (from LangChain):

PyPDFLoader: Extracts text from PDF files for indexing. Docx2txtLoader: Processes DOCX files and extracts their content. UnstructuredHTMLLoader: Parses HTML files to extract readable text.

Text Splitting:

LangChain RecursiveCharacterTextSplitter: Splits large documents into manageable chunks for embedding and retrieval, balancing context and performance.

Features

Document Upload: Users can upload PDF, DOCX, or HTML files via the Streamlit sidebar, which are then processed and indexed. Document Indexing: Uploaded documents are split into chunks, embedded using all-MiniLM-L6-v2, and stored in ChromaDB for retrieval. Chat Interface: A conversational interface where users can ask questions about uploaded documents and receive detailed responses. Retrieval-Augmented Generation (RAG): Combines document retrieval from ChromaDB with response generation via Gemini-1.5-Flash. Chat History: Persists and displays the conversation history for each session, stored in MongoDB. Model Selection: Allows users to choose the language model (currently Gemini-1.5-Flash) for response generation. Document Management: View a list of uploaded documents and delete them as needed through the sidebar.

Setup Instructions

Follow these steps to set up and run the Langchain RAG Chatbot locally. Prerequisites

Python 3.8+: Ensure you have a compatible Python version installed. MongoDB: A running MongoDB instance (local or cloud-based) for storing chat logs and metadata. Google Generative AI API Key: Obtain an API key from Google for accessing the Gemini-1.5-Flash model.

Installation

  1. Clone the Repository:
    git clone https://github.com/yourusername/RAG-VOICE.git
    cd RAG-VOICE
  1. Create and Activate a Virtual Environment:
python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate
  1. Install Dependencies:Install all required Python packages listed in requirements.txt:
pip install -r requirements.txt

Set Up Environment Variables:Create a .env file in the project root and add the following:

DB_URI: Your MongoDB connection string. DB_NAME: Name of the database to use. GEMINI_API_KEY: Your Google API key. API_URL: The FastAPI backend URL (default is local). RETRIEVER_K: Number of document chunks to retrieve (default is 5). LLM_TEMPERATURE: Controls the creativity of the language model (default is 0.7).

Running the Application

Start the FastAPI Backend:Launch the backend server in one terminal:

uvicorn src.main:app --reload

Start the Streamlit Frontend:In a separate terminal, activate the virtual environment again (if needed) and run:

streamlit run src/streamlit_app.py

Flowchart