BestEM Challenge - SQUAD CAL

Overview

Welcome to the BestEM Challenge repository. This project is a full-stack Retrieval-Augmented Generation (RAG) application designed to answer user queries by retrieving context from a specific knowledge base.

The system utilizes a pre-computed vector index (rag_index.pkl) to find relevant information and processes it through a backend server to generate accurate responses. It includes a Python-based backend, a web frontend, and standalone CLI scripts for testing and deployment.

---

Repository Structure

├── backend/            # Source code for the backend API logic
├── frontend/           # Source code for the web-based user interface
├── client.py           # A lightweight Python client for testing the API
├── server.py           # The main entry point for the backend server
├── rag_index.pkl       # Serialized vector store containing the knowledge base
├── start.sh            # Automation script to launch the full application
└── Presentation.pdf    # Project documentation and slides

---

Code Explanation

1. rag_index.pkl

This is a binary file created using Python's pickle module. It serves as the Knowledge Base for the RAG system. It contains:

• Vector Embeddings: Numerical representations of documents.
• Metadata: References to the original text chunks.
• Index Structure: (FAISS) enabling fast similarity searches.

2. server.py

The core backend logic resides here. Its primary responsibilities are:

• Initialization: Loads the rag_index.pkl into memory upon startup.
• API Management: Sets up endpoints (via FastAPI) to listen for incoming HTTP requests.
• Retrieval Logic: When a query is received, it searches the index for the most relevant documents.
• Generation: It combines the user query and the retrieved documents, sends them to an LLM (GPT4 in this case), and returns the generated answer.

3. client.py

A utility script used to test the backend without needing the frontend.

• It establishes a socket or HTTP connection to server.py.
• It sends a test payload (a sample question).
• It prints the raw response from the server to the console, useful for debugging API latency and payload structure.

4. start.sh

A Bash script designed to streamline the deployment process. It automates:

• Setting up environment variables.
• Launching the Python backend in the background.
• Starting the frontend development server.

---

Getting Started

Prerequisites

Before running the project, ensure you have the following installed:

• Python 3.8+
• Node.js & npm (Required for the frontend)
• API Keys: If the project uses OpenAI, ensure you have your API key ready.

Installation

1. Clone the Repository

   git clone [https://github.com/TeodorLepadatu/BestEM-challenge.git](https://github.com/TeodorLepadatu/BestEM-challenge.git)
   cd BestEM-challenge

2. Install Backend Dependencies Navigate to the .py files and install the Python requirements.

3. Install Frontend Dependencies

   cd frontend
   npm install
   cd ..

---

How to Run

Option 1: Quick Start (Recommended)

Use the provided shell script to spin up the entire environment at once.

chmod +x start.sh   # Ensure the script is executable
./start.sh

Option 2: Manual Execution

If you prefer to run components individually for debugging:

1. Start the Backend Server Open a terminal and run the server script.

# Ensure your API key is set if required
export OPENAI_API_KEY="your-api-key-here"

python server.py

Keep this terminal open.

2. Test with the Client (Optional) Open a second terminal to verify the server is responding correctly.

python client.py

3. Start the Frontend Open a third terminal to launch the web interface.

cd frontend
npm run dev   # or 'npm start', depending on the package.json scripts

Open your browser and navigate to the localhost URL provided (usually http://localhost:3000 or http://localhost:5173).

---

Troubleshooting

• ModuleNotFoundError: Run pip install <missing-package> (common packages used: flask, fastapi, langchain, openai, numpy, pandas).
• Pickle Load Error: If rag_index.pkl fails to load, ensure your Python version matches the version used to create the pickle file.
• Connection Refused: If client.py fails, ensure server.py is running and that the port numbers in both files match.