RAG FullStack Application

This is a complete Retrieval-Augmented Generation (RAG) application that uses a local DeepSeek model (via API), LangGraph for agentic workflows, and PostgreSQL (pgvector) for vector storage.

Features

• Full Stack RAG: End-to-end implementation from document ingestion to answer generation.
• Local/Remote Model Support: Configured to use DeepSeek API (compatible with OpenAI SDK).
• Vector Database: Uses PostgreSQL with pgvector for efficient similarity search.
• Agentic Workflow: Powered by LangGraph to handle complex queries and retrieval steps.
• Interactive UI: Built with Streamlit, featuring non-blocking background knowledge base updates.
• GPU Acceleration: Optimized for NVIDIA GPUs (CUDA 12.4) with CPU fallback.

Prerequisites

• Python 3.10+
• PostgreSQL with pgvector extension installed.
• NVIDIA GPU (Optional, but recommended for local embeddings).
• DeepSeek API Key (or compatible OpenAI-style API key).

Installation

1. Clone the repository (if applicable) or navigate to the project directory.

2. Install Dependencies:

   pip install -r requirements.txt

3. Environment Setup:

   • Ensure PostgreSQL is running.
   • Create a database (e.g., rag_db) and enable the vector extension:

     CREATE EXTENSION vector;

   • Set up your environment variables. You can export them in your shell or add them to the start.sh script.

Configuration

The application uses the following environment variables:

• HF_ENDPOINT: Mirror for Hugging Face (e.g., https://hf-mirror.com).
• DEEPSEEK_API_KEY: Your DeepSeek API key.
• DEEPSEEK_BASE_URL: Base URL for the API (e.g., https://api.deepseek.com).

Usage

One-Click Startup

Use the provided helper script to start the application:

bash start.sh

Manual Startup

1. Initialize/Update Knowledge Base: If this is your first run, or if you have added new documents to the data/ folder (or root), run:

   python create_db.py

   Note: You can also update the knowledge base directly from the Streamlit UI sidebar.

2. Run the Streamlit App:

   export HF_ENDPOINT=https://hf-mirror.com
   streamlit run app.py

Project Structure

• app.py: Main Streamlit application file. Handles UI and chat logic.
• create_db.py: Script to ingest documents and populate the vector database.
• rag_agent.py: Defines the LangGraph agent and retrieval logic.
• requirements.txt: Python dependencies.
• start.sh: Convenience script for launching the app.
• ARCHITECTURE.md: Detailed system architecture documentation.

Troubleshooting

• Torch/CUDA Errors: Ensure you have installed the correct version of PyTorch for your CUDA version.

  pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu124

• Database Connection: Check your PostgreSQL connection string in create_db.py and rag_agent.py.