Project Archive Answer Engine for Design and Engineering Firms

The problem

An architecture, engineering, or specialist contracting firm with twenty years of history has its most valuable asset locked in a network drive: past detail drawings, specification sections, RFI responses, submittal logs, cost histories, and site photographs. Nobody can find any of it. A question like which detail did we use for that parapet on a coastal site, or what did the engineer rule on that slab penetration in 2019, is answered by asking whichever senior person still remembers. When that person leaves, the knowledge leaves. Firms rebuild details they already own and repeat mistakes they already paid for.

Why now

Retrieval over messy technical archives became practical in the last eighteen months for three concrete reasons: long-context models can hold a full specification section and reason across it, vision-capable models can read title blocks, revision clouds, and dimensioned drawing sheets rather than only their text layer, and vector search infrastructure such as pgvector became cheap enough that a mid-size firm's entire archive costs a small monthly figure to index. Before that, drawings were effectively unsearchable and the whole idea collapsed at ingestion.

Who pays

Managing principals and IT leads at 20 to 300 person architecture practices, structural and MEP engineering consultancies, and specialist contractors in the US, UK, Canada, and Australia. They have a document management system such as Newforma, Procore, or Autodesk Construction Cloud, or just a well-organized file server, and an acute succession problem as senior staff retire.

How it makes money

Per-seat subscription of $40 to $90 per user per month with a floor, plus a one-time ingestion and mapping fee of $5,000 to $25,000 depending on archive size and messiness. Inference cost matters and must be priced in: embedding a large archive is a real one-off bill, and each answer costs a few cents in retrieval plus generation. At heavy usage that is single-digit dollars per user per month, which is fine against a $60 seat, but free-for-all usage on very long documents can erode it, so cap context and cache aggressively.

Market & demand

Order-of-magnitude: tens of thousands of AEC firms in the 20 to 300 person band across the four markets. A hundred firms averaging 40 seats at $60 a month is roughly $2.9M ARR before ingestion fees. This is a focused vertical SaaS, not a horizontal land grab.

AEC software is consolidating around Autodesk and Procore, both of which are adding assistant features. At the same time, the retirement of a large cohort of senior technical staff has made knowledge capture a board topic at mid-size practices. Firms are more willing to pay for something that reduces rework on details than for another collaboration tool.

SWOT

Competition & the gap

Newforma Project Center search, Autodesk Construction Cloud and Procore assistant features, Document Crunch and Trunk Tools in the construction document space, horizontal tools such as Glean, and the internal SharePoint search firms already dislike.

The wedge: Horizontal enterprise search does not understand a drawing sheet, a spec section number, or an RFI thread, and it does not know that the 2019 revision supersedes the 2018 one. Construction-specific tools focus on live projects and contracts, not the closed-project archive where the reusable technical knowledge sits. The gap is the archive, treated as a first-class asset, with citations a principal will accept.

Go-to-market

Sell one firm type in one region, starting with structural or MEP consultancies where reusable detail knowledge is densest. Lead with a paid archive audit that reports what is actually in their drive, which is valuable on its own and derisks the purchase.

First 10 customers: Use professional network introductions to three principals and offer a fixed-fee pilot: index one office's closed-project archive, run a workshop where senior staff ask their own hardest questions live, and convert on the result. Sponsor a regional professional institute CPD session on knowledge retention rather than buying ads.

How to set it up

1. 1Pick one discipline and interview five principals about the questions they actually ask the archive
2. 2Build ingestion for the messy reality: scanned PDFs, DWG exports, spec docs, email threads, photos, using Unstructured and a document OCR service
3. 3Implement retrieval with pgvector plus a reranking pass, and enforce that every answer cites document, sheet, and revision
4. 4Build a revision-awareness layer so superseded documents are flagged, not silently returned
5. 5Run one paid pilot with a workshop-based acceptance test
6. 6Complete a security questionnaire pack and per-tenant data isolation before the second customer

How to validate it

Daily active use by senior staff rather than only juniors, questions moving from search-like to judgment-like, pilots converting to annual contracts, firms asking to index a second office, and measurable reduction in redrawn details.

Tools & inspiration

Companies in this space: Newforma, Document Crunch, Trunk Tools, Glean, Procore

FAQ