RAG-FileQuery

This project implements a complete Retriever-Augmented Generation (RAG) pipeline using the Pathway framework. It indexes local documents (like PDFs and text files), generates embeddings, and uses Google's Gemini model to answer questions based on the content of those documents.

The application is architected as two microservices and is designed to run seamlessly with Docker Compose, providing a robust and reproducible environment.

The two-service architecture: A Flask web UI and a Pathway RAG API backend.

---

✨ Features

• Two-Service Architecture: A clean separation between the Flask web frontend (web) and the Pathway RAG backend (rag-api).
• Local Document Indexing: Processes files directly from a local data/ directory.
• PDF & Text Parsing: Automatically extracts text from various file formats, including PDFs (even scanned ones, with OCR).
• Advanced RAG Pipeline: Uses sentence-transformers for creating embeddings and pathway to build and serve a vector index.
• Gemini-Powered QA: Leverages a powerful Gemini model via litellm for intelligent question answering.
• Containerized & Reproducible: Runs with Docker Compose for a one-command setup that handles all dependencies, including system-level ones for OCR.
• Simple Web Interface: A clean web UI to ask questions and see the generated answers.

---

🛠️ Core Technologies

• Containerization: Docker, Docker Compose
• Backend Frameworks: Pathway, Flask, Gunicorn -e AI Engine: Google Gemini (via litellm)
• Embedding Model: sentence-transformers
• File Parsing: unstructured[pdf]

---

🚀 Getting Started with Docker (Recommended)

This is the simplest and most reliable method. It automatically handles all Python and system-level dependencies (like OCR tools) inside containers.

Prerequisites

• Git
• Docker and Docker Compose
• A Google Account to get an API key.

Step 1: Clone the Repository

git clone https://github.com/Per0x1de-1337/RAG-FileQuery.git
cd RAG-FileQuery

Step 2: Get Your Google API Key

1. Go to Google AI Studio.
2. Click "Get API key" and "Create API key in new project".
3. Copy the generated API key.

Step 3: Create and Configure the .env File

This file securely stores your API key and other configuration. Create a file named .env in the project's root directory and add the following content:

# .env file

# A strong, random secret key for Flask sessions
FLASK_SECRET_KEY='a-very-long-and-random-secret-string-for-security'

# This tells the web UI how to contact the RAG API inside the Docker network. Do not change.
RAG_API_URL=http://rag-api:8000

# Your Pathway license key (the demo key works fine)
PATHWAY_LICENSE_KEY="demo-license-key-with-telemetry"

# --- IMPORTANT ---
# Paste your Google API Key here
GEMINI_API_KEY="YOUR_API_KEY_HERE"

Note: The .gitignore file is configured to prevent .env from being committed to Git.

Step 4: Add Your Documents

Place the PDF and/or text files you want to query into the data/ directory.

Step 5: Build and Run!

With Docker running, use Docker Compose to build the images and start both services with a single command:

docker-compose up --build

• --build: Builds the Docker image using the Dockerfile. You only need to run this the first time or if you change requirements.txt.
• The first run may take a few minutes to download the base images and install dependencies.
• You will see logs from both the web and rag-api services.

Step 6: Use the Application

Once the services are running, open your web browser and navigate to: http://localhost:5000

You can now ask questions about the documents you placed in the data folder.

Stopping the Application

• To stop the application, press Ctrl + C in the terminal where Docker Compose is running.
• To remove the containers and network, run: docker-compose down.

---

▶️ Running with a Local Python Environment

This method requires manual installation of Python packages and system dependencies. It's more complex but useful for direct debugging.

Click to expand instructions for the local setup.

1. Prerequisites

• Python 3.10+
• A virtual environment tool like venv.
• System-level OCR dependencies for handling scanned PDFs.

On Debian/Ubuntu:

sudo apt-get update && sudo apt-get install -y tesseract-ocr poppler-utils

On Fedora/CentOS/RHEL:

sudo dnf install -y tesseract poppler-utils

On macOS (using Homebrew):

brew install tesseract poppler

2. Set Up Virtual Environment

# Create the virtual environment
python3 -m venv agent-venv

# Activate it
source agent-venv/bin/activate

3. Install Python Dependencies

pip install -r requirements.txt

4. Configure the .env File for Local Use

Create a .env file and use localhost for the API URL, as both servers will be running on your local machine.

# .env file (for LOCAL setup)
FLASK_SECRET_KEY='a-local-dev-secret-key'
RAG_API_URL=http://localhost:8000
PATHWAY_LICENSE_KEY="demo-license-key-with-telemetry"
GEMINI_API_KEY="YOUR_API_KEY_HERE"

5. Add Your Documents

Place your files into the data/ directory.

6. Run the Application

You need to run each service in a separate terminal.

In Terminal 1 (Start the RAG API):

# Make sure the venv is active
python3 app.py

Wait until you see the message ======== Running on http://0.0.0.0:8000 ========.

In Terminal 2 (Start the Web UI):

# Make sure the venv is active
python3 server.py

Wait until you see the message * Running on http://127.0.0.1:5000.

7. Use the Application

Open your web browser and navigate to http://localhost:5000.

---

📁 Project Structure

.
├── app.py                # The Pathway RAG API service (backend)
├── server.py             # The Flask web server (frontend)
├── docker-compose.yml    # Docker Compose orchestrator
├── Dockerfile            # Instructions to build the container image
├── config.yaml           # Configuration for the RAG service
├── requirements.txt      # Python dependencies
├── .env                  # (You create this) Stores secrets and API keys
├── data/                 # Folder for your input documents
├── templates/
│   └── index.html        # Web UI template
└── static/
    └── styles.css        # Web UI styles

---

🤝 Contributing

Contributions are welcome! If you have ideas for new features or improvements, please fork the repository and open a pull request.

---

📄 License

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