[05/08] Query — Question as Fingerprint

The question goes through the same translator

Before Search can rank anything, your question is turned into numbers — exactly like the PDF slices were. Same Embeddings step, same fingerprint idea. Model: the same embedding model used on the PDF slices.

How it works

Your question is plain text.
The same embedding model turns it into a fingerprint (8-D here).
Now it is directly comparable to the stored chunk fingerprints.

Type your question

Why this matters

“hospitals” and “Medical Center” become nearby numbers — so the next step (Search) can compare meanings, not letters. Without this fingerprint, Search has nothing to compare.

Try it — same meaning, different words

Type “Which medical centers are covered?” — the fingerprint stays close to the “hospitals” one. That’s the whole trick of vector search.

Next: Search

Now that the question is a fingerprint, Search ranks every chunk against it →