Feed It a Drawing.
Get Back a Bill of Materials.

Step-by-Step: From Drawing Submission to BOM

The drawing-to-BOM pipeline runs automatically once configured. Here is every step the system executes when a new drawing set arrives.

1. Drawing Ingestion. Files are received via email attachment, portal drag-and-drop, API POST, or connected cloud storage (SharePoint, Google Drive, Procore). The system accepts mixed sets — PDFs alongside DWGs alongside Revit exports — and normalizes them into a processing queue without requiring a consistent format from the submitting party.
2. Format Detection and Parsing. Each file is analyzed to determine its format and structure. Native CAD files (DWG, DXF, RVT, IFC, STEP, IGES, SolidWorks) are parsed with dedicated format readers that extract vector geometry directly — providing higher precision than raster image analysis for these types. PDFs are analyzed to determine whether they are vector (from a CAD export) or raster (scanned).
3. Sheet Organization and Indexing. Multi-sheet drawing sets are organized into a sheet index. The system reads title blocks to extract sheet number, title, scale, revision date, and discipline. Superseded sheets (same number, older revision date) are automatically flagged and excluded from extraction unless specifically requested.
4. Scale Calibration. Each sheet's drawing scale is established from the title block annotation and, where available, the scale bar graphic. For scanned drawings where scale bars may be distorted, the system uses multiple reference points to establish a calibrated scale. Scale calibration accuracy is logged per sheet and influences the confidence score for all linear and area quantities on that sheet.
5. AI Vision Extraction. The multimodal AI model processes each sheet to identify and count extractable items: symbols, component instances, linear runs, area fills, annotation callouts, and equipment tags. The AI draws on your configured knowledgebase to recognize project-specific symbols that may differ from standard drafting conventions.
6. Dimension and Quantity Computation. Raw detections are converted to quantities in your preferred units. Linear runs are summed. Areas are computed from boundary detection. Point symbols are counted. Annotations are parsed to extract explicitly stated quantities (e.g., "3-#4 rebar at 12" O.C.").
7. Specification Cross-Reference. Where specification sections are included in the submission (or previously loaded into the knowledgebase), each extracted item is matched to its governing specification section. Material grades, standards, and substitution clauses are read and appended to the BOM line item for use in pricing.
8. Confidence Scoring and Flagging. Each BOM line item receives a confidence score based on detection clarity, scale calibration quality, symbol recognition confidence, and specification match strength. Items below your configured threshold are flagged for human review. A review interface shows the source drawing coordinates for every flagged item.
9. Pricing Lookup and BOM Assembly. Extracted quantities are matched against your knowledgebase pricing data — supplier catalogs, labor rate tables, historical unit prices — and the BOM is assembled with extended costs. The final structured BOM is written to your workbook template and/or passed downstream to the proposal generation block.

Confidence Scoring: Trust but Verify

Every quantity extracted from a drawing carries a confidence score from 0 to 100. This score reflects the AI's certainty about the extraction, based on symbol clarity, annotation legibility, scale calibration quality, and knowledgebase match strength.

You set the confidence threshold for your workflow. Items above the threshold pass directly to the BOM. Items below the threshold are routed to a human review queue, where the reviewer sees the source drawing with the extraction highlighted — making it fast to confirm, correct, or reject.