Document RAG Assistant Suite

A suite of Streamlit applications that enable conversational interactions with document collections using Reranking Retrieval-Augmented Generation (RAG).

🌟 Overview

This project consists of two complementary Streamlit applications:

1. RAG_Bot - Upload and process new documents to create searchable knowledge bases.

2. Doc_Bot - Connect to existing document collections and query them conversationally.

Both applications utilize MongoDB Atlas Vector Search for document storage and retrieval, Cohere for embeddings, and Groq LLM for generating high-quality responses in French.

🚀 Features

RAG_Bot

• Upload various document types (PDF, TXT, CSV, DOCX)
• Automatic document processing and chunking
• Vector embedding generation with Cohere
• MongoDB Atlas Vector Search integration
• Real-time chat interface with the processed document
• Source citation and reference

Doc_Bot

• Connect to existing document collections in MongoDB
• Browse available knowledge bases
• Chat-based querying of document collections
• Source citation and reference
• Seamless integration with collections created by RAG_Bot

⚙️ Technical Stack

• Frontend: Streamlit
• Vector Database: MongoDB Atlas Vector Search
• Embeddings: Cohere (embed-multilingual-v3.0)
• LLM: Groq (meta-llama/llama-4-scout-17b-16e-instruct)
• RAG Framework: LangChain + LangGraph + Re-Ranking RAG
• Document Processing: LangChain document loaders

📋 Requirements

• Python 3.9+
• Streamlit
• MongoDB Atlas account
• Cohere API key
• Groq API key

🔧 Installation

1. Clone the repository:

git clone https://github.com/yourusername/doc-rag-assistant.git
cd doc-rag-assistant

2. Create a virtual environment and activate it:

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

3. Install the required packages:

pip install -r requirements.txt

4. Set up your .streamlit/secrets.toml file with your API keys:

MDB_URI_CONNECTION = "mongodb+srv://username:password@your-cluster.mongodb.net/?retryWrites=true&w=majority"
COHERE_API_KEY = "your-cohere-api-key"
GROQ_API_KEY = "your-groq-api-key"

🚀 Usage

Running RAG_Bot

streamlit run rag_bot.py

1. Upload a document using the file uploader in the sidebar
2. Click "Process Document" to create a new collection in MongoDB
3. Chat with your document using the chat interface
4. Note the collection name for future reference with Doc_Bot

Running Doc_Bot

streamlit run doc_bot.py

1. Select an existing collection from the dropdown in the sidebar
2. Click "Se connecter à la collection" to establish the connection
3. Chat with the selected document collection using the chat interface

🌐 Workflow

The typical workflow involves:

1. Using RAG_Bot to upload and process new documents, creating collections in MongoDB
2. Using Doc_Bot to access and query those collections later

This separation allows for a clean division between document processing and querying functionality.

🔍 How It Works

1. Document Processing:

   • Documents are loaded using LangChain document loaders
   • Text is split into manageable chunks with RecursiveCharacterTextSplitter
   • Chunks are embedded using Cohere's multilingual embedding model
   • Embeddings are stored in MongoDB Atlas with vector search capabilities

2. Retrieval Process:

   • User questions are embedded using the same Cohere model
   • MongoDB Atlas Vector Search finds semantically similar document chunks
   • Retrieved chunks provide context for the LLM

3. Response Generation:

   • LangGraph orchestrates the retrieval and generation flow
   • Groq LLM (LLama 4) generates responses based on retrieved context
   • Responses are provided in French with proper source citations

📝 Notes

• All responses are generated in French as specified in the prompt template
• Document collections are stored in the "IT_OPS" database in MongoDB Atlas
• Collections created by RAG_Bot are prefixed with "doc_" and include a timestamp
• Source references are provided with each response

⚠️ Requirements File

Create a requirements.txt file with the following dependencies:

streamlit
pymongo
langchain
langchain-core
langchain-mongodb
langchain-cohere
langchain-groq
langchain-community
langgraph
python-dotenv

🛠️ Configuration

Ensure your MongoDB Atlas cluster has:

1. Vector search capability enabled
2. Proper network access settings
3. A user with read/write permissions

👥 Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.