Labelling

Input requirement

Labelling currently requires a PowerPoint .pptx file containing labelled diagrams. It does not accept a standalone image through the SuperAgent route. Slides without a usable embedded diagram or without real printed labels are skipped rather than filled with invented labels.

The workflow

1. Read the deck. KoiSwarm inspects the embedded images on each slide and ignores images repeated across several slides, such as logos and templates.
2. Choose the main diagram. The largest suitable embedded image on a candidate slide is processed.
3. Read printed labels. Vision OCR finds label text and its real position. A slide with no detectable printed labels is not turned into a quiz.
4. Identify structures. The pipeline associates each detected label with the structure it names and drops low-confidence results.
5. Hide the answer text. The printed labels are covered while the surrounding diagram remains visible.
6. Build questions. Each accepted structure receives one correct answer and three distractors. Questions are ordered by their position on the diagram.
7. Open the practice page. The generated slide image and questions are stored for the interactive Labelling surface.

Better decks

Use a PPTX whose diagrams are embedded as clear images. Printed labels need to be legible, and the main diagram should not be a low-resolution thumbnail. Slides may be skipped when labels overlap, OCR cannot find dependable text, or the structure cannot be identified confidently.

Check the generated key

Labelling is active-recall practice, not an authoritative anatomy reference. Compare ambiguous structures with the original slide and your course material before relying on the answer key.