README_en.md

Paper-Chat

Paper-Chat is an AI chatbot that enables conversations about academic papers using arXiv IDs.
It is designed to assist researchers and students in quick learning, efficient practical application, and trend analysis, helping them easily understand and explore complex academic content.

Prerequisites

1. Docker and Docker Compose must be installed.

Installation and Execution

Production

1. Clone the repository

   git clone https://github.com/alchemine/paper-chat.git
   cd paper-chat

2. Build the cluster (entrypoint, elasticsearch cluster) and run the Streamlit app using Docker Compose

   docker-compose -f docker-compose.prd.yml up

3. Access the Streamlit app in your browser

   http://localhost:8501

Development

1. Clone the repository

   git clone https://github.com/alchemine/paper-chat.git
   cd paper-chat

2. Build the cluster (entrypoint, elasticsearch cluster)

   docker-compose -f docker-compose.dev.yml up

   • Otherwise, use dev container

3. If you want to use AzureChatOpenAI, required environment variables must be set in the dev.env file.

   OPENAI_API_KEY=...
   
   # Needed if using Azure LLM
   AZURE_OPENAI_ENDPOINT=...
   AZURE_OPENAI_API_KEY=...
   OPENAI_API_VERSION=...
   AZURE_OPENAI_LLM_DEPLOYMENT_NAME=...
   AZURE_OPENAI_LLM_MODEL=...
   AZURE_OPENAI_EMBEDDINGS_DEPLOYMENT=...
   AZURE_OPENAI_EMBEDDINGS_MODEL=...

   • If you write OPENAI_API_KEY, OpenAI API Key value is automatically filled with the value.
   • You can use AzureChatOpenAI with azure input in OpenAI API Key.

How to Use

1. Enter the OpenAI API Key and the arXiv ID of the paper in the sidebar.

• The paper ID is in the format like 2004.07606.
• If you input a string that can identify the ID, such as https://arxiv.org/pdf/2004.07606 or https://arxiv.org/abs/2004.07606, the ID will be automatically identified.

2. A summary of the paper is automatically generated and displayed.

• When generating a summary for a paper for the first time, it takes about 30 seconds.
• Subsequent requests will load the previously generated summary.

3. You can ask questions about the paper through the chatbot interface and receive answers based on the paper's content.

The queries used for exploration (Queries) and the referenced content (Contexts) are displayed along with the answer. Check these to see if any hallucination has occurred.

4. If you want to ask questions about a different paper, enter a new arXiv ID in the sidebar.

Additional Features and Improvements

Please refer to the Issues page for major development progress and updates.