Example 1 — a dimension question, answered and cited
"What is the loading dock leveler pit depth?"
"The loading dock leveler pit depth is 11 inches, measured from finished floor
[RFI-2026-042, Sheet T-1, Rev A]."One sentence, one number, one checkable source — sheet and revision included. Anyone on the job can verify it in fifteen seconds. That's the standard every answer is held to: no citation, no claim.
Example 2 — the one it refused to guess
"Per the attached Addendum A3, what is the fire rating of mechanical room door M-101? Please cite the source."
"The sources don't specify the fire rating of door M-101…" — followed by which document would resolve it.
This is the answer we're proudest of. A generic chatbot would have produced a confident, plausible, invented fire rating — on a life-safety question. IntelMS read the document, found the rating genuinely absent, and said so. A missing-document answer is a useful answer; a guessed one is a hazard. More on that: what's safe to use AI for on construction documents.
Example 3 — revision-aware by default
"What is the latest RFI log revision and its date? Please cite the source."
"The latest RFI log with an explicit revision and date is the one marked Revision A, dated 2026-06-08…" — with the source document named.
Asked at 2:41pm, answered at 2:42pm. The system also distinguishes a document's stated revision from its file metadata — the kind of trap that catches humans working fast. Why that matters: the real cost of building from a superseded revision.
The guardrails behind these answers
Every answer above passed the same gate: grounded only in the project's uploaded documents, latest revision, exact source cited, conflicts flagged instead of silently resolved — and anything contractual, cost, code, or safety-critical routed to a human. A person approves replies; the system just makes them fast and defensible. The full record of every Q&A lands in a per-project timeline for closeout and dispute defense.
Run this test on your own project
14-day pilot on one real project. Email your documents in, ask the questions you argued about last week, and judge the citations yourself.
Start a free pilot