We structure TIFF scanned drawings with a deterministic × AI hybrid. Not estimates — takeoff you can measure to the millimeter.

TIFF scanned drawings hold only “pixels” — machines cannot understand lines, text, or components. Relying on a VLM alone makes dimensions mere “estimates,” and miscounts and hallucinations propagate into the estimate (= cost).

• We want to convert pixel-only TIFFs into structured data

  Machines cannot understand lines, text, or components
• We want to measure dimensions precisely and count 100% of quantities

  A VLM alone yields estimates, miscounts, and hallucinations
• We don’t want errors propagating into the estimate (cost)

  Estimates amplify errors via unit price × quantity
• We want to trace and audit “why this value”

  A black box cannot be explained