Metal Fabrication
Every request for quote arrived with technical drawings that estimators had to interpret by hand: identifying every part, quantity, material, and dimension before pricing could even begin. We built an agentic AI product that extracts this information from drawings, gives estimators a structured review experience, and prepares the inputs needed to build a quote.
Users
Metal estimators
Before
5+ hrs per job
After
auto-extracted parts table
When a customer requested a quote, they sent an email with technical drawings and supporting files. An estimator had to separate the fabrication drawings from unrelated attachments, review each one individually, identify every unique part, and extract its quantity, material, dimensions, and specifications.
That information then had to be transferred into an estimation spreadsheet, where the estimator calculated material weight and cost, determined the right fabrication machines and labor, and assembled everything into a final quote.
The estimator's expertise was needed for pricing and production decisions, not repetitive transcription from drawings.
The estimator uploads a drawing package, or forwards the request-for-quote email to a shared inbox. The system identifies the relevant drawings, extracts each fabrication item, and creates an editable takeoff table: mark ID, quantity, material grade, description, thickness, length, width, and cutouts. The estimator validates the information, corrects values, and adds anything the system missed.
AI interpreted the drawings. Deterministic shop rules calculated the quote: material weight and cost, machine selection, and labor.
01
A customer's quote email and its attachments are captured under a unique quote record.
02
The system filters PDF and image attachments, separating drawings from unrelated documents.
03
Drawings are classified as simple parts or complex assemblies and processed accordingly.
04
The system identifies parts and extracts quantities, materials, dimensions, and specifications.
05
The estimator reviews an editable table, corrects values, and adds missing parts before approving.
06
Validated takeoff data feeds material, machine, labor, and pricing calculations.
01
Customer emails mixed drawings with unrelated files. The system classified attachments first, so downstream agents only processed documents relevant to the takeoff.
02
A flat plate and a welded, bent assembly need different interpretation. Complexity classification let the workflow treat them differently instead of forcing every drawing through one process.
03
Extracted values were a starting point, not a final answer. Estimators could correct fields, add missing parts, and override the system before anything entered the quote.
A takeoff identifies what needs to be manufactured. A quote also has to determine what it will take to manufacture it. Once the estimator approves the extracted data, the product calculates material area and weight, cost, adjusted margin, labor and machine requirements, shipping inputs, and the final estimated quote value, using shop-specific rules to route each part to the right cutting equipment.
The product connected what the drawing specified to how the part would be manufactured and priced.