We use haystack to search the documents
Currently there is a script, which splits the response from e.g. https://cms.integreat-app.de/api/muenchen/de/pages into different documents. It also removes the HTML tags. These generated files can easily be converted and stored by haystack document.
To run this script use python3 parse-files.py
To run the first prototype you need to download a LLM. There are different available on huggingface. For example google-flan-t5-xl
Before running the script some dependencies need to be installed, therefore create a virtual environment and activate it. The script will download the llm and it will be saved in home/.cache/huggingface, so expect the first run of the script to be significantly slower than the following ones.
python3 -m venv .venv
source .venv/bin/activate
then install
pip install --upgrade pip
pip install pip install farm-haystack==1.17.2