lumos-visualization-llm-nli

Lumos: Enhancing Data Visualizations with LLM-based Natural Language Interfaces

Setup

Prerequisites

You will need Docker (Docker Engine & Docker Compose) and an OpenAI API key.

Setup Steps

1. Create a .env file containing the OpenAI API key stored in the variable OPENAI_API_KEY
2. Open terminal or powershell
3. Build container
   1. Windows: docker-compose build
   2. Linux: docker compose build
4. Run system
   1. Windows: docker-compose up
   2. Linux: docker compose up
5. Initialize Backend
   1. cd into scripts folder and run npm install
   2. npm run datasets to import datasets
   3. npm run usecase to import use case llm
6. Open Frontend

VLAT Tests

If you want to run tests on your own:

1. cd into scripts folder
2. npm run tests to import tests
3. npm run runtests to run ALL tests. This may require a high balance on your OpenAI account. Alternatively you can execute single tests in the UI Tests section
4. npm run extract to extract all tests / test results to a json file

Notes

When you interact with the chat, the message thread is persisted in the OpenAI backend, but not yet in the Lumos system. This means if you re-open the frontend, there is a blank chat window. But behind the scenes there is still your old messages on OpenAI. So the LLM might return confusing information, like control information containing a different color scheme from a previous session. For a fresh thread, you need to remove the LLM and the related prompt via API calls.