Processing pipeline
When you upload a drawing, Bedrock’s vision pipeline processes it through several stages:- Drawing Preprocess — Split the PDF into individual pages, creating one Sheet per page.
- Sheet Preprocess — Segment each Sheet into Blocks. Detect block boundaries, classify types (Plan, Schedule, Legend, etc.), and extract title block metadata (sheet number, title, discipline).
- Block Preprocess — Process each Block based on its type. For view blocks (Plan, Elevation, Section, Detail): detect Features, extract grid lines, identify cross-references. For text blocks (Notes, Schedules): run OCR and extract structured content.