Information collection for transparency story

[schroedk]

We want to showcase an example to fulfill the transparency requirements from the AI act. There is a first collection in the docs.

Ideas:

• json file for static system information (e.g. intended use, responsible persons/institution, contacts)
• model signature (source: mlflow) + model prediction endpoint (swagger UI)
• evaluation results (metrics + plots)
• fairness scores with explanation
• data source
• huggingface model card template

Missing information (not yet traced):

• commit hash
• serving endpoint runtime (dependencies) information (make available via separate endpoint), check this collection from clearml for inspiration
• implement raw information collection step in pipeline (result zip archive logged in datalake/ mlflow)

Things to check: