How do we make and validate machine learning concepts without the most basic raw materials? How can we effectively capture and create data values ​​when resources are scarce?