Construction plan search AI: ask the set in plain language, get the current detail

Keyword-searching a folder of PDFs finds you files. It does not find you the answer. The fire rating on the corridor wall is on a sheet somewhere, and the product that achieves it is in a spec section somewhere else, and the field needs both before the inspector shows up. Knowledge workers spend close to a fifth of the week just hunting for and gathering information.^[1] In a 300-page set on a phone at the wall, that fraction goes up. Brad reads the drawings and specs and lets you ask them by what they actually say.

The cost of all that hunting is not abstract. US construction pros lose roughly 14 hours a week to non-optimal work, about 5.5 of them spent looking for project data and another 3.9 on rework that often traces back to acting on the wrong version of a detail.^[2] A plan question answered in ten seconds, cited to the current sheet, is time the schedule gets back. A plan question answered from memory, or from a sheet that got superseded two revisions ago, is how the rework starts.

Search by content, not by file name

A file name tells you a sheet exists. It does not tell you what is on it. Brad reads the drawings and the specs, the sheets and details and schedules and CSI MasterFormat sections, so you can ask in plain language instead of guessing which PDF to open. “What is the fire rating on the corridor wall?” comes back with the rating and the sheet it lives on, not a list of files that might contain it.

That means no memorizing the sheet index, and no pinch-zooming a 300-page set on a phone mid-install. The question is the search. You ask the way you would ask the person who drew it, and the answer arrives with the page attached.

Plans and specs, read together

A real answer usually straddles both documents. The drawing shows the assembly; the spec section defines the product that goes into it. Ask about a wall type and the real answer is the detail plus the governing CSI section, read as one thing. Brad connects them, so a question about a detail can pull the spec section that controls it right alongside, instead of leaving you to find the second half yourself.

The link matters most when a revision moves one document and forgets the other. A detail gets updated, the spec stays as it was, and now the set quietly disagrees with itself. Brad flags that drift instead of returning whichever sheet it happened to read first. A discrepancy between the drawings and the spec is a human's call, and the contract defines the formal process for resolving it; Brad's job is to surface the conflict early, not to paper over it with a confident guess.^[3]

Every answer carries its source

The field acts on an answer it trusts and ignores one it cannot place. So Brad returns the sheet number, the revision, and a link to the exact spot, every time. A super confirms the spec at the wall without a call to the office. An estimator traces a quantity back to the page it came off, instead of carrying a number with no provenance into a schedule of values. The citation is not decoration. It is what turns a plausible answer into one someone will pour concrete on.

Revision is the part people skip and then regret. An answer off the right sheet at the wrong revision is worse than no answer, because it carries false confidence. Brad pins each answer to the current sheet and flags when the detail you are looking at has been superseded. That is the difference between an as-built that matches the building and a closeout package that needs an apology.

Fast enough for the field

The whole point is the ten-second answer in the moment it is needed: at the wall, on the phone, mid-install. Not a callback an hour later, by which time the crew has moved on or guessed. A guess at the wall is how a detail gets built wrong, and a wrong detail is rework, the single largest avoidable line in the productivity tax. Direct rework runs around five percent of total construction cost on its own.^[4]

Ask Brad over the channels your team already uses, by text or email, and the answer comes back with the page attached. There is no new app to open at the wall and no sheet index to scroll. The crew asks, the project answers, and the install keeps moving. The slow version of this, where the question waits for someone back at the trailer to find the sheet, is exactly the waiting that lean production treats as non-value-adding waste.^[5]

Where Brad stops and the design professional starts

Brad is document intelligence pointed at construction drawings. It reads what you give it and shows its work. It does not invent a detail to fill a gap in the set, and it is not a stand-in for the design professional whose stamp owns the answer. When the documents genuinely disagree, Brad flags the conflict and points at both sources rather than resolving it on your behalf. Your project's content stays yours, and each workspace is walled off from every other one. If you have specific requirements about how a set gets handled or retained, ask us and we will walk you through exactly how it works.

The answer to the plan question is almost always already in the set. The only question is whether the field can lay hands on the current one before it has to act. Brad reads your plans and specs, answers in plain language, and attaches the sheet, so the answer arrives with its source instead of as one more thing to verify later.