Cadastre
Half a century of records, asked a question.
A family-owned commercial real estate firm with over fifty years of records sitting across paper, microfiche, early spreadsheets, and modern systems. We built Cadastre, a pipeline that digitised what had never been digitised and gave the family one place to ask their own history a question.
Fifty years of leases, surveys, maintenance records, and correspondence sat in every format a long-running business accumulates. Paper in filing cabinets, microfiche, early spreadsheets, modern PDFs. Three family members carried most of it in their heads. The data was always somewhere. It was never anywhere they could ask.
Cadastre is built in two halves. The first is a digitisation pipeline with a custom OCR layer tuned for the quirks of fifty years of paperwork, hand-written addenda, faded carbon copies, inconsistent date formats, addresses recorded six different ways. Everything is normalised into a typed schema. The second half is a retrieval-augmented generation layer, with every record embedded as multimodal vectors and indexed in a vector database. The family asks the archive questions in plain language and gets answers with citations back to the original document, scanned or otherwise. The system runs continuously. Fifty years of fragmented memory became one query box.
More Work
Explore other projects from our portfolio
