Skip to content
SheetMetalAI: illustrative product screen

SheetMetalAI

AI ductwork takeoff and quoting.

folders
Construction & Property
client
SheetMetalAI
date
November 2026
Live product
sheetmetalai.com

SheetMetalAI makes takeoff and quoting software for HVAC sheet metal shops in the US. Vizio AI's Venture Studio designed and built the web app and the AI pipeline behind it. An estimator uploads a mechanical PDF, checks the ductwork the models find, and sends a quote priced with the shop's own rates.

Product strategy
AI & LLM
Web development

THE PROBLEM

The bid is due Friday and every duct is still uncounted.

Before a sheet metal shop can price a job, someone has to count and measure every duct, elbow, damper, tap and flex box on the mechanical plans. Estimators do it sheet by sheet, and one plan set can take days.

Those days decide how many jobs a shop can bid. The people behind SheetMetalAI are HVAC contractors, and they wanted a tool that does the first count, prices it the way their shop prices, and leaves the final call to the estimator.

WHY IT STILL MATTERS IN 2026

Mechanical bids still arrive as PDF plan sets, and manual takeoff still caps how many bids a shop can send each week. Rival tools sell on accuracy percentages a buyer cannot check, so a takeoff the estimator can see and correct on the plan is the honest pitch.

Source: Vizio AI review of HVAC takeoff software sites

WHAT WE BUILT

From mechanical PDF to priced quote.

Five AI stages turn a plan set into a takeoff, and the estimator checks the result before a single dollar is added. A vision language model reads the sheets, a segmentation model traces the ductwork, and a second pass reads sizes and links every part. The shop's own price book does the pricing.

A vision language model finds the ductwork plans and their scale on each PDF sheet, a tiled segmentation model traces every duct and fitting, a second model pass reads the size labels and links connected parts, the estimator corrects the takeoff in the plan editor, and the shop's own price book turns it into a PDF quote.

INSIDE THE PRODUCT

Inside the takeoff.

Real screens from the app, running on a sample office plan set.

Takeoff
01 · TakeoffThe AI finds every duct on the plan and colours it by part type.
Quote overview
02 · Quote overviewParts are priced from the shop's own price book into one total.
Projects
03 · ProjectsEach project keeps its client, plans and quoted price.
Quoted value
04 · Quoted valueQuoted value by month, so the owner sees the pipeline grow.

HOW WE WORKED

Plan in, quote out in 17 weeks.

We started with the hardest part, reading a plan, and had the upload talking to the vision pipeline in the first days. The editor, accounts, billing and pricing followed until the loop from PDF to quote worked. Since launch the models and the product have kept improving: tiled multi-scale detection, flex box detection, a per-class accuracy gate on every model change, labor priced by the pound and client records.

Timeline: the build starts on day 1, AI detections appear on the plan in week 2, the zoomable plan editor lands in week 8, accounts and a test environment in week 11, billing and the price book in week 15, and the full PDF to quote loop goes live in week 17.

DESIGN

Built for an estimator who checks every count.

A takeoff the estimator cannot check is a takeoff they will not send, so every step shows its work on the plan. These loops run the real app on a sample plan set.

Plan editor · AI detections ready for review
Quote page · takeoff priced with shop rates
Dashboard · projects, plans and quoted value

The estimator stays in charge

Every detection can be resized, retyped or removed on the plan before anything is priced. Sizes the model cannot read are flagged instead of guessed.

The shop's own numbers

Each shop keeps its own price book, flex box rules, sketching hours and labor by the pound. Prices freeze per project, so an old quote never shifts.

Output a customer can read

One download holds the PDF quote, with weights, liner and totals, plus the marked-up plan images behind it.

OUTCOME

What changed.

90%
Less estimator time per takeoff
3x
More bids sent per estimator each week
324
Parts counted and priced on a three-sheet sample job

Any trade that still prices work by counting items on drawings, such as electrical, plumbing or fire protection, fits the same pattern: a model does the first count, the expert corrects it on the drawing, and their own price book turns it into a quote.

START A PROJECT

Still counting items on drawings by hand?

Tell us what your estimators do today. We reply with how we would build it and a rough timeline.

What do you want to build?

From the case study: SheetMetalAI

FAQ

About this project.

01

02

03

04

More works.

Bancroft Estimating: illustrative product screen
Bancroft Estimating
Construction Cost Estimating Client Portal with Online Payments

A website and secure client portal for a construction cost-estimating business.

Power BI Real Estate Sales, Telesales & Lead Conversion Dashboard: illustrative product screen
Power BI Real Estate Sales, Telesales & Lead Conversion Dashboard
Construction & Property · Confidential client

A family of reports covering sales, the telesales team and leads, including project, unit, bank and channel breakdowns.

Automated Lease Contract Tracking & QuickBooks Invoice Reconciliation: illustrative product screen
Automated Lease Contract Tracking & QuickBooks Invoice Reconciliation
Construction & Property · Confidential client

A background system that reads lease contracts from Drive, turns them into building, unit, tenant and lease records, and checks them against QuickBooks invoices.

Think AI-first.
Move future-fast.

Let’s build what’s next! From product ideas and intelligent workflows to data-driven systems and scalable growth operations.

Book a Strategy Call