An RFI, a request for information, is the formal question a contractor
sends the design team mid-construction when the drawings disagree or fall short. Work on
that item waits until the answer comes back. Most RFIs exist because the documents
contradict themselves, and those contradictions are findable before the set is issued.
796
RFIs on the average project
9.7 days
median wait for an answer
That is about $860,000 of answering questions on a single project.
Source: Navigant Construction Forum.
Why technology can help
Reading every sheet against every other one by hand takes days, so the checks that
would catch these problems almost never happen before a set goes out. Technology can do
it in seconds, in two ways that matter equally.
An LLM reads an entire set at once and
surfaces the contradictions and gaps that would otherwise come back as RFIs. Every
quote it cites is verified against the drawing before you see it.
Automated checks pulled straight from the
PDF catch the mechanical errors that slip past a tired pair of eyes: references to
missing sheets, callouts to details never drawn, marks absent from the schedules.
We built a working demo of both. Try it on a real bid set, then see
the full methodology for how each check works.
Where this goes
This is an early build, but it already works on real bid sets. It gets much more useful
with a company's own data:
- The norms we hardcoded into the checks could instead be learned from a firm's own
standards.
- The contradiction model could be tuned on the RFIs a firm has already answered, so
it catches the problems that actually recur.