From uploaded pages to editable questions
Preserve page coordinates, group questions and align answers so every item can be checked against its source.
Start with the structure of the page
Teaching documents mix text, formulas, tables and figures. Flattening a page into plain text can lose the relationships between a question, its choices and its diagram. Vispo starts with positioned OCR elements that support grouping and source crops.
The current code defaults to a dots.ocr / dots.mocr integration. This is an open model we integrate. Vispo's engineering connects its output to question structure and teacher review; we do not claim authorship of the foundation model.
Grouping questions and aligning answers
Text and coordinates are organized into question groups, response types, source crops and answers. Questions and solutions may come from different pages, so alignment retains numbering, provenance and structure for teacher review.
Keep uncertainty reviewable
Completeness checks identify items that still need attention. When a check fails, the pipeline preserves the source crops and marks the review requirement, keeping uncertainty visible.
This describes the inspected code integration. Served model versions depend on deployment configuration. Upstream benchmark results are not Vispo product results.