Construction Plan Archive Search
Construction Plan Archive Search lets a BIM Manager or design-tech lead ask plain-English questions across the firm’s full archive of project drawing sets in PDF, DWG, and IFC, and get citation-grounded answers tied to specific sheets and bounding boxes. It is available now as a Labs engagement under MSA + Appendix, built on the same Document AI core used in our Document AI Agent.
- Semantic search across the firm’s archive — calibrated per deployment to the customer’s storage volume (typically tens-to-hundreds of thousands of PDF pages) — with citation back to the source page.
- Plain-English questions resolve to drawing-content matches, not filename matches, so “stair detail in healthcare projects” finds the right sheets even when filenames are generic.
- Indexes vector PDFs, scanned drawings via OCR, DWGs, and IFC exports so legacy CAD, BIM exports, and redlined record sets are all queryable.
From “where did we draw that?” to “show me the sheet”.
Workflow today
-
01
Senior architect needs a precedent — Day 0
A senior architect remembers a clean stair detail from a healthcare project around 2022 and asks the team to “find that detail”. Nobody recalls the exact project number, phase, or whether the final version lived in Revit, DWG, or a PDF issue set.
Day 0 -
02
Junior staff trawls the archive — Day 1–3
A junior architect or BIM coordinator spends days in SharePoint, BIM 360 / ACC, and old file servers. They guess at filenames, open random PDFs, and skim 30–50 sheets per set because nothing in the folder structure encodes “stair detail, healthcare, 2022”.
Day 1–3 -
03
Best-effort match found — Day 3+
After several days, the team finds a detail that is approximately right, or they give up and redraw from memory. Past decisions in Revit, DWG, and PDF never connect to current work, and the same stair, curtain wall, or podium waterproofing detail is redrawn project after project.
Day 3+ -
04
Knowledge stays trapped — ongoing
The firm’s 10-year archive across CAD, BIM, and PDF is effectively a black box. No one can ask “which projects used this curtain wall typology?” or “where did we resolve this MEP clearance?” in a structured way, so knowledge from coordination and mechanical drawing PDF parsing never compounds forward.
Ongoing
Workflow with VitruAI
-
01
Index the archive — Week 1, one-time
As a Labs engagement, VitruAI configures the Construction Plan Archive Search workflow on top of the Document AI Agent. The agent ingests PDFs (including scans), DWGs, and IFC exports from file shares, BIM 360 / ACC, and DMS exports into a structured corpus. It reads sheet titles, view types, callout references, room tags, and key text so “A-501 stair detail 12” becomes findable by concept, not just by name.
Week 1, one-time -
02
Ask the question — Day 1, ~1 min
The senior architect types a plain-English query such as “show me all stair details from healthcare projects in the last 5 years” or “projects that used the same curtain wall typology as Tower X”. The workflow uses the same semantic understanding as our window and door takeoff from PDF pipeline, tuned for precedent search instead of quantity extraction.
Day 1, ~1 min -
03
Review cited answers — Day 1, ~5 min
Within single-digit minutes, the agent returns a ranked list of drawing pages with thumbnails, sheet numbers, project names, and phases. Each hit highlights the bounding box where the relevant stair, curtain wall, or mechanical detail lives, and stores that interaction in the firm’s Memory Agent so repeated queries get faster and more targeted over time.
Day 1, ~5 min -
04
Open the drawing — Day 1
From the result list, the architect jumps directly to the source PDF page, DWG layout, or IFC-exported view in the project archive. Details are reused instead of redrawn, precedent sets are assembled in ~30 min instead of days, and knowledge from prior mechanical drawing PDF parsing or coordination cycles becomes a living reference library.
Day 1
Questions about Construction Plan Archive Search
-
How big can the archive be?
Archive size is calibrated per deployment; Labs engagements typically index tens-to-hundreds of thousands of PDF pages plus associated DWG and IFC exports over a 1–2-week ingest window. Index time scales with file count and storage location, but query response sits in single-digit seconds once the corpus is built. The index can be extended incrementally as new projects close out or as legacy projects are backfilled from BIM 360 / ACC.
-
Where does the index live?
For Labs work, the default is a VitruAI-hosted index with the firm’s source documents remaining in their existing locations such as file servers, SharePoint, or BIM 360 / ACC. Some firms prefer self-hosted or hybrid deployments, which we scope case by case under MSA + Appendix. In all cases, the index is a read-only derivative of PDFs, DWGs, and IFC exports, and the firm can request deletion or re-ingest at any time.
-
Does it index DWG and IFC, or just PDFs?
Construction Plan Archive Search reads vector PDFs, scanned PDFs via OCR, native DWG, and IFC exports with format-aware parsers. The same extraction stack that powers window and door takeoff from PDF and mechanical drawing PDF parsing is extended to support sheet titles, view names, callouts, and keynotes for precedent search. Mixed archives where some projects are only in DWG and others only in IFC are supported as part of the Labs scoping.
-
What about confidentiality across projects?
Query-time access controls mirror the firm’s existing project-permissions model, using roles and groups from systems such as BIM 360 / ACC or SharePoint where available. A user only sees search results from projects they already have permission to view, so confidential work or restricted pursuits do not bleed into general precedent search. For highly sensitive projects, the firm can exclude them entirely from the Construction Plan Archive Search corpus during the Labs configuration.