cgrevisse/raspatientpi

RasPatient Pi: A Low-cost Customizable LLM-based Virtual Standardized Patient Simulator

RasPatient Pi is a low-cost customizable LLM-based virtual standardized patient simulator. The simulator leverages automatic speech recognition, LLMs and text-to-speech engines. Scenarios can be specified by the teacher through a short description, while the clinical knowledge of LLMs is used to coherently complete any gap in the scenario.

The software is implemented as a Django-based web application. It uses OpenAI's API under the hood.

The usage of OpenAI's API will generate costs. Users of RasPatient Pi are solely responsible for any generated cost and should closely monitor their usage!

Installation

After cloning or downloading the code, install the dependencies using PIP:

virtualenv env
source env/bin/activate
pip install -r requirements.txt

Create the database tables (we are using the default SQLite configuration, feel free to use any other database engine):

python manage.py makemigrations
python manage.py migrate

The software uses MQTT for asynchronous message exchange and web sockets. Install a broker such as Mosquitto:

sudo apt install mosquitto

Use the provided mosquitto.conf file to start the broker:

/usr/local/sbin/mosquitto -c mosquitto.conf

If Mosquitto is automatically started on your platform, create a link to the configuration:

sudo ln -s mosquitto.conf /etc/mosquitto/conf.d/mosquitto.conf

Make sure to allow port 9001 in your firewall:

sudo ufw allow 9001

Configuration

Copy the template config-dist.py to create the configuration file:

cp config-dist.py config.py

Create a DJANGO_SECRET_KEY and fill in your OPENAI_API_KEY. Before the first run, create an Assistant:

python manage.py createassistant

Deployment

RasPatient Pi can be deployed on a single-board computer (such as a Raspberry Pi) to be used alongside a manikin or played in a browser, relying on a 3D avatar.

On our Raspberry Pi, we had to install a bunch of packages to get things running:

sudo apt install libcairo2-dev pkg-config python3-dev libgirepository1.0-dev portaudio19-dev python3-pyaudio firmware-sof-signed libgstreamer1.0-0 gstreamer1.0-dev gstreamer1.0-tools libx264-dev libjpeg-dev libgstreamer1.0-dev libgstreamer-plugins-base1.0-dev libgstreamer-plugins-bad1.0-dev gstreamer1.0-plugins-ugly gstreamer1.0-tools gstreamer1.0-gl gstreamer1.0-gtk3 gstreamer1.0-qt5 gstreamer1.0-pulseaudio

To check configure the sound, use alsamixer.

Finally, to start RasPatient Pi, use:

python manage.py runserver

Citation

@inproceedings{RasPatientPi2024,
    author={Grévisse, Christian},
	title={{RasPatient Pi: A Low-cost Customizable LLM-based Virtual Standardized Patient Simulator}},
	booktitle={Applied Informatics},
	year={2024},
}