01The drawing set

Hundreds of sheets, one register

A live construction project runs to hundreds of drawings, and every sheet carries the same facts in its title box: the number, the revision, the title, the scale, the date and who issued it. Those facts have to reach a register, because the register is how anyone knows whether the sheet on site is the current one. Schedules carry even more: a door schedule or a finishes schedule is a table inside a PDF, often spread over several pages, holding the detail that pricing and procurement both depend on.

Traditionally all of it is retyped by hand, again on every revision. The work is slow, done by people who are expensive to have doing it, and it is where errors enter a project quietly. A mistyped revision does not announce itself until something is built wrong.

02What we built

Reading the set automatically

Agency AI builds the AI inside APSIS Business Components, a construction software platform. Two of the capabilities we have built read the drawing set itself.

Drawing extraction takes a batch of drawings, reads those details off each sheet, and produces a register with every field populated and every sheet traceable back to the file it came from. It handles every drawing you give it in one go, because drawings never arrive one at a time.

Schedule extraction pulls the tables out. You tell it which pages to look at. A schedule sitting on pages 40 to 47 of a long document is pulled out without touching the rest, and what comes back is data rather than a picture of a table.

03Accountability

Every field carries its history

On a live project this data feeds procurement and pricing, so what happens when our AI gets a field wrong matters as much as how often it gets one right. Our AI’s answer is never the final word. Every field can be corrected, and when someone corrects one the system records what the model said, what the value was before, what it was changed to, and who changed it.

A reviewer can see at a glance which fields a person has touched and which stand as the model left them. A dispute about where a number came from has an answer. And because corrections cluster where the model is weak, the record shows exactly where to improve it. The same pattern runs through every document capability in the platform: extraction, comparison and analysis all keep the model value beside the human value rather than overwriting it.

Drawing register · extracted fields
Sheet A-204 — ground floor planREV CEXTRACTED · 1 FIELD CORRECTED
01Drawing number — A-204As extractedModel value
02Revision — CWas B, corrected by a reviewerEdited · recorded
03Scale — 1:100As extractedModel value
04Issue dateAs extractedModel value
Illustrative. Corrections sit beside the model value rather than replacing it.
04The value

Hours saved, errors prevented

The obvious saving is time. Two hundred drawings is a day of somebody’s week, repeated on every update, so a project loses weeks of skilled time to typing. The larger saving is the errors that never happen: a hand-typed register drifts out of date almost immediately, and it becomes serious when material is ordered against the wrong revision, or when a claim depends on proving which sheet was current in March.

Once the register is live data rather than a spreadsheet somebody maintains, new questions become routine. What has been superseded and not reissued. Which schedules moved between revisions. Which sheets have never been picked up on site.

05Where it leads

Feeding the rest of the platform

Extracted data earns its keep downstream. A structured schedule can be priced. A populated register can be checked against what is on site. A readable document can be questioned. These capabilities share their projects, permissions, document handling and audit trail with estimating and quoting and procurement document processing, all built with APSIS over more than a year.

Read how we build the AI inside the platform →

06Built for production

Designed for how drawings actually arrive

Three design choices carry it in daily use. It takes the whole set at once, because drawings arrive as issues rather than single sheets. It takes a page range, so a schedule inside a two-hundred-page document does not require processing the lot. And extraction runs as a task with a status a team can watch, retry and assign, rather than a button that either works or silently does not.

07What it does

How it works in practice.

Send the lotGive it every drawing at once and it reads them all, rather than one at a time.
Just the pages you needTell it the schedule is on pages 40 to 47 and it ignores the other 150.
Nothing gets overwrittenIf someone changes what our AI read, the system keeps both versions and who changed it.
Next

Start with one process.

Tell us where your team loses the most time. We will tell you honestly whether AI pays there, what it takes to build, and what we have already delivered for businesses like yours.