Every roofing bid starts with a takeoff. Find the roof page, check the scale, then measure every plane, hip and valley. Skilled estimators do it well, but it’s slow, and it happens one file at a time.
For Superior Roofing, we built Roof Pro, a dimension extractor: upload an architectural PDF and get a measured roof, a full takeoff summary and a branded quote — with a model that learns from every job.
The takeoff
The estimator uploads a blueprint set. The platform reads the title block for the address and builder, finds the roof plan page in a multi-page set, and cross-checks the scale across pages to catch inconsistencies before any measuring starts. A crop tool isolates the roof area, which makes extraction faster and more accurate.
From there, it calculates surface area, edge lengths and wastage. Scale and line extraction use deterministic logic, so the measurements are repeatable. AI handles the judgment calls: what each line is, and which lines belong together.
Impact on operations
Before, takeoffs were slow and manual, and the team could only work through one estimate at a time. Now Superior’s estimators work in a shared, AI-assisted platform that handles multiple estimates at once, running on real production files with 80–90% usable accuracy.
Every completed estimate also becomes training data. A built-in shred clock tracks time per estimate to support SR&ED documentation. What used to be a one-off task is now building a proprietary roofing AI asset that belongs to Superior.
Enter Roof Pro
Roof Pro is built on the Claude API and runs on AWS inside Superior’s own environment. It pairs Claude with a custom classification model trained on roofing-specific line types.
What Roof Pro does:
→ Reads the plans, finding the right page, validating scale and pulling project details automatically.
→ Classifies every roof line as hip, gable, valley, ridge or eave, joining overlapping and fragmented lines into single elements.
→ Builds the quote, turning the takeoff into a branded PDF ready to send to the customer.
Why the estimator stays in charge
A bid is a promise to a customer, so a measurement is only useful if the estimator trusts it. We built Roof Pro around a human-in-the-loop review. Estimators accept or reject each roof face with clear visual indicators, and can add their own line categories and colours for edge cases.
Nothing goes out without an estimator’s sign-off. Their corrections feed the training pipeline, so the system keeps improving because the team uses it.
“This is what I was envisioning from the beginning.”
— Superior Roofing
