Elastic Chat

Elastic Chat is an chat application that integrates Elasticsearch with GPT-3.5 for efficient information retrieval and conversation.

Project Structure

Elastic Chat is organized into several key components:

elastic-chat/
├── app/
│   ├── config.yaml
│   ├── chat.py
│   ├── run_web.py
│   └── search_query.py
├── create-embeddings/
│   ├── models/ (gitignored)
│   ├── pdf/ (gitignored)
│   ├── load_embeddings_model.py
│   └── load_pdf_docs.py
├── .env (gitignored)
├── .gitignore
├── environment
└── README.md

Key Components

• app/: Contains the core application scripts for chat operations and configurations.
  • config.yaml: Configuration file for setting up user, model, system prompts, and responses.
  • chat.py: Handles chat interactions, including generating responses using GPT-3.5.
  • run_web.py: A script to run a GUI for interactive chatting.
  • search_query.py: Manages connections to Elasticsearch and performs search queries.
• create-embeddings/: Scripts for processing documents and loading models to Elasticsearch.
  • load_embeddings_model.py: Loads distilroberta model into Elasticsearch for enhanced search capabilities.
  • load_pdf_docs.py: Parses PDF documents and indexes their content into Elasticsearch.
• .env: A gitignored file for storing environment variables such as Elasticsearch credentials.
• environment: Defines the project's environment setup and dependencies.

Setup Instructions

1. Environment Setup: Setup a conda environment by using en. Refer to the environment.yml file. (load_embeddings_model.py has a separated environment.yml because of incompatibilities with libraries but it don't cause incovenience in the deployment)

2. Configuration:

   • Create a .env file in the root directory of the project.

   • Populate the .env file with your Elasticsearch and OpenAI credentials as follows:

     ELASTIC_CLOUD_ID="your_elastic_cloud_id_here"
     ELASTIC_USERNAME="your_elastic_username_here"
     ELASTIC_PASSWORD="your_elastic_password_here"
     ELASTIC_INDEX="your_elastic_index_here"
     OPENAI_API_KEY="your_openai_api_key_here"
     

   • These credentials are essential for connecting to your Elasticsearch cluster and for using the GPT-3.5 API. Ensure you replace the placeholder values with your actual credentials.

3. Elasticsearch Model: Before running the chat application, ensure you've loaded the dense vector model into Elasticsearch using load_embeddings_model.py for optimal search performance.

   • Create your deployment with a machine learning run instance and with an index with the name you entered in your .env
   • You can crawl information from a web creating web-crawler and activate a pipeline to create a dense vector representation using the loaded model.
   • For this step you can help you with this blog post

4. Running the Application:

   • To index documents into Elasticsearch, use the load_pdf_docs.py script from the create-embeddings directory.
   • • example command: python create-embeddings/load_pdf_docs.py --index_name <your_index_name> --pdf_path <path_to_your_pdf.pdf>
   • To start the chat GUI, run run_web.py with the command streamlit run run_web.py from the app directory. This will open a web interface for interactive chatting. (Created simply with streamlit)