Work Done

• Determined a way to measure feature granularity / overlap as we go down
• Separated and cleaned up embedding generation
• Claude Project Feature Granularity

Confusions

• What does each metric I made mean?
• Do they actually measure how broad each concept becomes

Next Steps

• Find the most similar vectors from one layer to the other and see what happens when you apply them??

• Separate out results as well because some datasets are overpowering others.

• Analyze results and see if I can come to a conclusion
• If so then try to invalidate those results
• Try other methods of validation as well

Cool Resource

• Mastering Cursor