Cut drawing intake from two hours to ten minutes per project for a 30-person engineering firm
Client snapshot
A 30-person engineering firm that prices and plans fabrication work from clients' drawings. Each project arrives as a set of CAD-exported PDFs plus instructions in Word documents and email threads.
The problem
Someone had to read every drawing and type what mattered into the estimating sheet: part references, dimensions, materials, notes and revision numbers. Then they read the Word documents and the email thread for instructions that changed what was on the drawing. A typical project took about two hours of intake before any engineering started. Retyping errors found their way into estimates a few times a month, and the firm only found out when the parts were wrong.
What we built
An intake folder and a pipeline behind it.
When a project folder lands in Google Drive, n8n picks it up. CAD-exported PDFs are vector files, so a Python step with PyMuPDF reads them directly: every text block with its position, the line and shape geometry, and the CAD layers they came from. No OCR, so dimensions come out exactly as drawn. Claude then groups the text into the fields the estimators use, matching notes to the parts they sit next to.
Word documents and emails go through the same step. Claude pulls out the instructions, such as "use 316 stainless on all brackets" or "ignore revision B", and links each one to the drawing it changes.
Everything is written to structured JSON and a row per part in Airtable, with a link back to the page it came from. Each field has a confidence flag. Anything below the bar, or any conflict between a drawing and an instruction, goes to a needs-review view instead of straight into the estimate. We wrote the acceptance criteria with the firm before building and tested against twenty past projects.
Results, 60 days in
- Intake per project: about two hours → about ten minutes, mostly checking the review view.
- Around 90% of fields extracted with no person involved.
- Retyping errors reaching estimates: several a month → close to none.
- Conflicts between drawings and instructions now show up at intake, not on the shop floor.
Quote
"The review view is the part I didn't know I needed. It shows me the three things on a drawing that don't agree with the email." Lead estimator
Build snapshot
- Timeline: 7 business days
- Stack: n8n, Python, PyMuPDF, Claude, Google Drive, Airtable
- Investment: $2,200
- Who owned it: the lead estimator set the fields and the review thresholds and clears the review view daily.
The pattern
This is the pattern we ship in the Document Sorter engagement. Here the documents were engineering drawings; the same build reads invoices, contracts or intake forms, with your fields swapped in.
This is an illustrative composite drawn from operator engagements. Names, identifying details, and exact metrics are anonymised. The workflow, stack, and approach are real.
Looks like your situation?
A 30-minute call tells us whether the same pattern fits your operation, and what it would cost to ship it.