This is a proof-of-concept application meant to demonstrate a compentece search / guidance tool based on generativ AI using the MRST database of papers and documentation.

  git clone https://github.com/aasmund-mjos-sintef/mrst-rag-project.git

• You can use the .env.example file as an example. Any environment variable associated with LangSmith is not needed, but can help with debugging openai_api errors or find out where and why the program does something stupid.
• It's important to download the original MRST repository if you want to use the git search. If you don't download the original MRST repository, you will get an error if you include git search in the settings. You can go to mrst github to download it.

The essential stuff is

  LANGCHAIN_OPENAI_API_KEY = <your_openai_api_key>  
  MRST_REPOSITORY_PATH = <full_path_to_downloaded_mrst_repository>

You need to be on python version newer or equal to 3.13

  # Create a virtual environment in the .venv directory
  python3 -m venv .venv
  # Activate the virtual environment
  source .venv/bin/activate

For this project you need graphviz downloaded on your computer:

• On Debian/Ubuntu systems run

  sudo apt install graphviz

• On Mac, you can use homebrew

  brew install graphviz

  pip install -e .

To run the program, navigate to the frontend folder and enter

  streamlit run app.py

If everything is set up correctly, you should see the image below. However while the source code is being imported, the buttons to the right will not be visible. The first time you run the program,this will probably take a while. This is because the vector embedding models have to be downloaded on your computer. If you haven't specified the path to the mrst repository in the .env file, make sure the Git button is switched off.

For a detailed list of the articles included in the datasets, please go to src/mrst_competence_query/datasets and check out the two .txt files for a detailed list