We took a real construction project's records, more than 800 pages of specs, contracts, and submittals, plus over 300 pages of drawings, and turned them into one system the team can ask a plain-English question and get a cited answer from.
The problem
On a project, the answer to a single question is scattered across specs, drawings, submittals, RFIs, and contracts, each in its own place. Finding one requirement means digging through hundreds or thousands of pages by hand. And the drawings, where much of the real detail lives, are the hardest thing of all to search.
What we built
Working from a real project's document set, we built the system end to end:
- Ingested 800+ pages of text documents, specs, contracts, submittals, chunking, embedding, and indexing every page so it can be searched by meaning, not just exact keywords.
- Processed 300+ pages of construction drawings through a vision pipeline that converts each sheet, plans, sections, schedules, into searchable text, so the drawings join the same searchable set instead of being left out.
- Built the query interface where the team asks a question in plain English and searches across the entire project's documents, with every answer cited back to its source.
The system surfaces and cites; the people decide.
The result
One place to ask a whole project's documents a question and get an answer tied to its source, text and drawings together, instead of hunting page by page across separate systems. The knowledge that was buried across hundreds of pages becomes something the team can actually find.
