How search over your documents works

Move the sliders and watch a search pipeline decide what the model is allowed to see — and what happens when it is fed the wrong passage.

Chunk size 8
Overlap 0%
Score threshold 0.70

Tap a part of the diagram to open that step.

The whole pipeline

Document → pieces with overlap → numbers → filter by fields → nearest by distance → cut off by threshold → what is left goes to the model. Nearly every wrong answer in such a system is decided before the model is even called.

What's inside

  1. 1 Cutting the document up — Watch the red marks: those are boundaries that cut a sentence in half.
  2. 2 Nearest by meaning — Score is a distance, not a fact check.
  3. 3 Filtering before the search — Turn the filter off and watch the adult row climb back into the results.
  4. 4 The threshold and the empty answer — Nothing found is a valid outcome. A model with no passages should refuse, not improvise.
  5. 5 What reaches the model — The answer never tells you which passage it came from. Only the pipeline above decides that.

Explorable Explanations

Chunk size 8
Overlap 0%
Score threshold 0.70
1 / 5