Does this work on scanned drawings?

Yes, the workflow supports scanned drawing PDFs using an OCR pipeline tuned for architectural linework and text. Accuracy depends on scan resolution, contrast, and how clean the originals are, so 300 dpi or better and minimal compression artefacts give the best results. Handwritten markups and very faint text are not guessed; they are flagged for QS review with a clear link back to the affected sheet region.

How accurate are the takeoffs?

Accuracy for window and door takeoff from PDF depends on drawing quality, symbol standards, and how consistently type marks are used across the set. Each Labs engagement ships a per-project accuracy report calibrated to the customer’s own drawings and a ground-truth sample graded by the QS team. Low-confidence extractions are highlighted for manual confirmation, so the workflow never silently asserts counts where the underlying PDF is ambiguous.

Can it handle drawings from older archives?

Yes, legacy scanned drawings from archives can be processed after an initial calibration pass. The Labs engagement includes a review of historical title blocks, symbol conventions, and common annotation styles so the extractor can recognise older lineweights and text fonts. For broader archive questions beyond openings—such as locating specific details across decades of projects—you can pair this with construction plan archive search in the same Document AI stack.

How does this fit with the parent Document AI Agent?

Window and Door Takeoff from PDF is one specialised workflow within the Document AI Agent, focused on opening schedules when a Revit model is not available or not contractual. The same agent family also handles mechanical drawing interpretation, including mechanical drawing PDF parsing, archive-search use cases, and contract-document extraction. Firms typically calibrate several of these workflows together so that document-side extraction complements model-side tools like BIM-to-BOQ automation.

Use Case VitruAI Labs

Window and Door Takeoff from PDF

Window and Door Takeoff from PDF extracts opening schedules from architectural drawing PDFs—type marks, dimensions, quantities, handing, glazing—without requiring a Revit source model. It reads vector and scanned sheet sets, identifies door and window symbols, and emits a structured CSV or Excel file with sheet and bounding-box citations, available now as a Labs engagement under MSA + Appendix.

Door and window schedules extracted from drawing PDFs, calibrated per deployment, typically in single-digit minutes per typical project sheet set.
Every schedule line in the CSV ties back to the source sheet number and bounding box on the page for quick visual verification.
Works on vector PDFs and scanned drawings via an OCR pipeline, with low-confidence items flagged for review instead of guessed.

Scope a Labs engagement See capabilities ↓

How it works

From measure-by-eye to read-the-PDF

Workflow today

01

Drawing PDFs land in QS inbox

Day 0. The architect issues a drawing set as PDFs; no Revit or IFC model is shared, or the model is not contractual. The QS receives mixed sheets—plans, elevations, schedules, and details—with no machine-readable opening schedule and no link to tools like the BOQ Takeoff Agent.

Day 0
02

QS counts openings by hand

Day 1–4. The QS steps through each plan, elevation, and window or door schedule sheet, counting symbols by eye and marking tallies on printouts or in Bluebeam. Type marks are transcribed manually, dimensions are re-keyed, and opening types are inferred from tags or legends, with errors compounding quietly across dozens of sheets.

Day 1–4
03

Schedule built in Excel

Day 4–5. Counts are re-entered into Excel or Google Sheets, grouped by door and window type. Handing, fire rating, glazing, and hardware prep are copied from notes or legends. Discrepancies between elevations, plans, and any partial schedules are caught late, and there is no direct audit trail back to the original drawing PDF for each row.

Day 4–5
04

Pricing waits

Day 5+. The estimator cannot finalise pricing until the opening schedule stabilises. If the architect issues a revised PDF set, the QS repeats large parts of the manual takeoff. On projects where a Revit model exists, it still may not be shared, so BIM-to-BOQ automation cannot be used for this tender.

Day 5+

Workflow with VitruAI

01

Upload the PDF set to the agent

Day 1, ~10 min. The QS uploads the architectural drawing PDFs—plans, elevations, sections, and any existing schedules—to the same Document AI stack that powers the Document AI Agent. The workflow classifies sheets into architectural, schedules, and details, and prepares both vector content and scanned pages for extraction.

Day 1, ~10 min
02

Agent extracts the schedule

Day 1, ~1 hour. The workflow detects door and window symbols, reads type marks, and parses dimensions and annotations directly from the drawings. It builds an opening schedule with quantities by type, handing, glazing, and any clearly tagged performance attributes, and ties every row to the originating sheet and bounding box so a QS can click back to the exact region in the PDF.

Day 1, ~1 hour
03

QS reviews flagged items

Day 1–2. Low-confidence items—faint scans, overlapping markups, or unusual type-mark conventions—are flagged for human review instead of silently accepted. The QS steps through these cases in a review UI, correcting type marks or counts where needed, similar to how they would validate mechanical drawing PDF parsing outputs for MEP openings.

Day 1–2
04

Export to Excel / CSV

Day 2. Once reviewed, the QS exports the window and door schedule to CSV or Excel for pricing, or passes it to the BOQ Takeoff Agent to connect with downstream quantity and rate workflows. For projects that later move into model-based workflows, the extracted schedule can sit alongside BIM-to-BOQ automation as the PDF-side equivalent.

Day 2

Common questions