Search with AT2 Citations

An application mimicking LLM-powered search engines with citations provided by AT2. Given a search query, this application

1. Invokes the Tavily API to retrieve relevant web pages.
2. Uses an LLM (by default, Microsoft's Phi-4-mini-instruct) to respond to the query given information from the relevant web pages.
3. Provides citations, i.e., references to the part of the web pages used by the model, for any part of the response.

search_with_at2_citations.mov

Prerequisites

Running this application requires the following:

• Node.js
• npm
• Python
• Tavily API key for search
• GPU with CUDA support for running the LLM

Installation

Clone the repository

git clone https://github.com/MadryLab/search-with-at2-citations.git
cd search-with-at2-citations

Backend Setup

1. Navigate to the backend directory:

cd backend

2. Install the required Python packages:

pip install -r requirements.txt

3. Set your Tavily API key by adding the following to your .bashrc or equivalent:

export TAVILY_API_KEY="your_tavily_api_key_here"

Frontend Setup

1. Navigate to the frontend directory:

cd ../frontend

2. Install the required Node.js packages:

npm install

Running the Application

You can run both the backend and frontend with the run.sh script provided in the root directory:

./run.sh

The frontend will be available at: http://localhost:3000