Module 09 Summary

What this module established

An embedding is a direction; only comparisons between directions mean anything. 'Similar' means similar in whatever the model was trained to distinguish - inspect what your near-neighbours share before trusting it.

Carry forward

• Nearest-neighbour search always returns k results. Add a threshold, and read the rank-1-to-rank-2 gap as a measure of decisiveness.
• Detect near-duplicates by embedding similarity before splitting. File hashes miss video frames and burst photos entirely.
• Record which model produced every index. Mixing models produces plausible, meaningless rankings with no error anywhere.

Before moving on

Move on when your duplicate check ran before splitting and your index records which model built it.

Mark complete