A diligence workspace that shows its evidence
A real estate investment company was running acquisitions out of shared drives and mail threads. We built them a private workspace that reads every document in a deal, answers only from those documents, and points at the exact page or row behind each answer.
- IndustryReal estate investment
- EngagementInternal tool, designed and built by us
- DeploymentPrivate, on the client's own hardware
Built for the whole arc of a deal
Every property gets its own managed folder. Whatever lands in it is indexed and filed by type, and the workspace asks for confirmation instead of filing a document silently.
From there the deal carries a status through sourcing, diligence, offer, closing and into ownership, and the workspace can say what a file is still missing. One place holds the whole arc.
Deal knowledge sat in files nobody could search
The documents that decide a purchase were arriving faster than anyone could file, read or check them.
The evidence was everywhere at once
Listings, valuations, title and land registry papers, tax records, community documents and correspondence arrived by mail and settled into shared drives. Answering one question meant remembering which file held it.
Summaries nobody could check
A written summary is worth exactly what the reader's trust in it is worth. With no line back to the source page, every conclusion had to be re-derived by hand before anyone would act on it.
Watching the market by hand
Stock a buyer was waiting on had to be re-checked manually, and an asking price that moved between two checks left no trace. Nobody could see how a price got to where it stood.
A workspace that answers from the file and shows you the file
Each capability exists because a person has to be able to check what the system said.
Answers with the page and the row behind them
Ask a question in plain language and the workspace answers from that deal's own documents and nothing else. Every claim carries a citation you can open.
- Citations that resolve to the exact page of a PDF
- Spreadsheet answers cited down to the individual row
- Click a citation and land on the highlighted source
Commentary that has to evidence itself
Generated market writing is checked before it can be published. Language that promises certainty, predicts a sale price or reads as investment advice is rejected outright, in every language the system writes.
- Promises of certainty and predicted prices fail the check
- Every published note must point at the evidence behind it
- The same check runs before sending and again on arrival
It still answers when the model is down
List and table questions are answered straight from the indexed records with no model in the path, so the workspace keeps working through an outage.
- List and table questions answered without a model
- Written for how people actually type, across several languages
- A question with no match comes back empty
Runs on your own hardware and stays there
The workspace, its index and its model all run on the client's own machine. Sending anything to a cloud model is a deliberate setting, never a default.
- Documents are indexed locally and stay local
- In the packaged build only the application reaches the model
- Uploaded text is cleaned before it ever reaches the model
What changed in the work
No figures here. The client's numbers are the client's, and we do not publish a percentage we cannot show the working for.
Checkable
Every answer
A reviewer opens the cited page and confirms the claim at its source, instead of re-reading the file it came from.
One place
The whole deal
Sourcing, diligence, offer, closing and ownership sit in the same workspace, and each property carries what it is still missing.
Your hardware
Where it runs
The documents, the index and the model stay on the client's own machine, and reaching for the cloud is a decision someone has to make on purpose.
Does your work turn on what is in the file?
If the answer lives inside documents, we build the system that reads them and shows you where every answer came from.
Discuss your project