25 years of drawings, searchable in plain English.
Staff at an architecture firm were losing around an hour each a week hunting through job folders. We built a search that answers questions the way a colleague would, with links back to the source files, and the drawings never leave the firm's own server.
Context
The practice has around 50 people: architects, technologists, associates, directors and admin. Twenty-five years of work sits on a Windows file server, about 32TB of it, filed in job folders by the year each project started.
The problem
Finding anything meant knowing which job it was filed under. Staff spent around an hour each a week searching, and many simply gave up and started again on work that already existed somewhere. New joiners needed about three months before they could find things on their own. And the question asked most often was the hardest to answer: which house types have we done in this style, in this local authority, at this size?
What we built
A plain-English search over the whole archive. Staff type the question as they would ask a colleague and get an answer with citations back to the original drawings and documents. Text is extracted and indexed on the firm's own hardware, and personal mailboxes are deliberately left out.
- 01Reads drawings, PDFs and project folders on the file server
- 02Extracts and indexes the text on the firm's own hardware
- 03Answers plain-English questions with links to the source files
- 04Keeps excluded folders out of everyone's results
The decision that mattered
We kept the files on the firm's own server. The obvious build would have sent every drawing to cloud services for parsing and indexing, which means years of confidential client work leaving the building, and a cloud bill on top. Doing that work locally kept the drawings in-house and the running costs down.
of drawings and project files indexed
of archive searchable in one place
staff asking questions in plain English