Code for "Conversations at Scale: Robust AI-led Interviews with a Simple Open-Source Platform"

There are two options to explore the AI-led interviews discussed in the paper.

Option 1: Online notebook

To try own ideas for interviews within minutes and without the need to install Python, see https://colab.research.google.com/drive/1sYl2BMiZACrOMlyASuT-bghCwS5FxHSZ (requires to obtain an API key)

Option 2: Full platform

To install Python and set up the full interview platform locally (takes around 1h from scratch), see the following steps.

The interview platform is built using the library streamlit and the APIs of OpenAI and Anthropic.

• Download miniconda from https://docs.anaconda.com/miniconda/miniconda-install/ and install it (skip if conda is already installed)
• Obtain an API key from https://platform.openai.com/ or https://www.anthropic.com/api. In case of the OpenAI API, choose a "project" key
• Download this repository
• In the repository folder on your computer, paste your API key into the file /code/.streamlit/secrets.toml (requires to make hidden folders visible)
• In the config.py, select a language model and adjust the interview outline
• In Terminal (Mac) or Anaconda Prompt (Windows), navigate to the folder code with cd (if unclear, briefly look up basic Linux command line syntax for navigating to folders)
• Once in the code folder, create the environment from the .yml file by writing conda env create -f interviewsenv.yml and confirming with enter (this installs Python and all libraries necessary to run the platform; only needs to be done once)
• Activate the environment with conda activate interviews
• Start the platform with streamlit run interview.py

Paper and citation

The paper is available at https://ssrn.com/abstract=4974382 and can be cited with the following bibtex entry:

@article{geieckejaravel2024,
  title={Conversations at Scale: Robust AI-led Interviews with a Simple Open-Source Platform},
  author={Geiecke, Friedrich and Jaravel, Xavier},
  url={https://ssrn.com/abstract=4974382},
  year={2024}
}