Vector Search with LLM in InterSystems IRIS

Overview

This project demonstrates a Vector Search application using Large Language Models (LLMs) integrated with InterSystems IRIS. The application is built using Streamlit and leverages OpenAI’s API for natural language processing tasks.

Project Structure

├── app
│   ├── docker-compose.yaml
│   └── streamlit
│       ├── combined_code.py
│       ├── constants.py
│       ├── db_operations.py
│       ├── Dockerfile
│       ├── helper.py
│       ├── main.py
│       ├── openai_llm.py
│       └── requirements.txt
├── jupyter_notebooks
│   ├── intersystems_vector_search.ipynb
│   └── sample_data
│       ├── example_Imatinib_Teva.pdf
│       ├── large_document.txt
│       └── small_data.txt
└── README.md

Prerequisites

Ensure you have the following installed on your local machine:

• Docker
• Docker Compose

Getting Started

• Clone the Repository
• git clone https://github.com/mychoys-co/intersystems_vector_search
• cd app

Set Up Environment Variables

Create a .env file in the root directory of your project and add the following environment variables:

IRIS_USERNAME=demo
IRIS_PASSWORD=demo
IRIS_HOSTNAME=intersystems
IRIS_PORT=1972
IRIS_TABLE_NAME=intersystems_table
IRIS_NAMESPACE=USER
OPENAI_API_KEY=<your-openai-api-key>

Build and Run the Application

If Using the app through docker and UI

Run the following command to build and start the services:

docker compose up --build

If encounter docker timeout issue do

docker compose down
docker compose up --build

If want to use the app through jupyter notebook

Run Intersystems DB

docker run --name iris-comm -p 1972:1972 -p 52773:52773 -e IRIS_PASSWORD=demo -e IRIS_USERNAME=demo intersystemsdc/iris-community:latest

Run jupyter notebook

cd jupyter_notebooks
jupyter notebook

This will start two services:

• InterSystems IRIS: Runs on ports 1972 and 52773.
• Streamlit App: Runs on ports 8501 and 8502.

Access the Application if using through docker

Once the services are up and running, you can access the Streamlit application by navigating to:
http://localhost:8501

Usage

Uploading Files

You can upload a .pdf or .txt file through the Streamlit interface.
The file content will be processed and inserted into the InterSystems IRIS database.

Asking Questions

Enter your question in the provided input box and click the Submit button.
The application will generate a response to your question using OpenAI’s API and display relevant search results from the InterSystems IRIS database.

Closing app

If using though docker compose

Simply do Ctrl+C from the terminal

docker compose down

If using jupyter notebook

• Close notebooks
• Close Intersystems, do Ctrl+C

docker rm iris-comm

Project Components

docker-compose.yaml

Defines the services required for the application: intersystems (IRIS) and app (Streamlit application).

streamlit/Dockerfile

Specifies the Docker image for the Streamlit application, including the installation of necessary dependencies and models.

streamlit/main.py

Contains the main Streamlit application code, which handles file uploads, data processing, and querying.

streamlit/db_operations.py

Handles database operations such as verifying table existence, inserting data, and counting records.

streamlit/combined_code.py

Combines various functionalities such as initializing the database, inserting data, and generating responses to user queries.

streamlit/openai_llm.py

Handles interactions with OpenAI’s API for natural language processing tasks.

streamlit/requirements.txt

Lists the Python dependencies required for the Streamlit application.

streamlit/constants.py and streamlit/helper.py

Contain various helper functions and constants used throughout the application.

Notes

Ensure that the InterSystems IRIS container is fully up and running before attempting to use the Streamlit application.
The sleep 25 command in the Docker configuration ensures that the IRIS service has enough time to start before the application attempts to connect.

Troubleshooting

If you encounter any issues, check the logs of both containers using the following command:
docker-compose logs