The approach used is based on the CRFill Model, which can be found here.
To use the model, we need to simply call process.py in the docker environment (the container must supply the /input and /output folders), following exactly the Node21 setup.
All the code for running it (including a trained model) is contained in the model_submission folder.
The complete code setup can be found in the crfill folder, in which there is also an example script in the readme of how it was trained.