Most organisations hold three versions of every vendor, customer and asset. We set the strategy, build one golden record, and keep it clean — so agents act on facts, not guesses.
Before any agent goes live, we agree which data matters, who is accountable for it, where it must live, and which parts a model may see. The result is a plan your board and your regulator can read.
| Domain | Owner | May AI use it? | Lives in |
|---|---|---|---|
| Vendors | Head of Procurement | Yes, bank fields masked | Dhaka |
| Production | Plant manager | Yes | Dhaka |
| Customers | Head of Sales | Rules and lookups only | Dhaka |
| Financials | Chief Financial Officer | Summaries only | Dhaka |
| Employee records | Head of HR | No | Dhaka |
A master data model gives every system, app and agent the same definition of a property, a supplier or an employee. It is in production today: a commercial real-estate model for Menkes, version 2.7.9, with 155,000 reference rows, bound to Groundfloor.
Entity layout shown here is an example. Names and screens are illustrative.
Every field has one meaning, one format and one owner — in English and, where needed, Bangla.
On Groundfloor, the model feeds the apps teams use daily: asset inventory, property forms, people and roles, a task inbox.
Changes to structure are proposed, approved by a person, and recorded, so you can see what changed and why.
Rules find likely duplicates. A data steward sees both records side by side, with the reason for the match, and decides. Nothing changes the master record without sign-off.
Our File Content Extractor turns PDFs, scans, Word files, spreadsheets, CSV, text, audio and video into text, with an optional model pass for messy pages. Then rules check the result against your master data. Extracted is not the same as trusted.
Dataplane is one governed layer over your databases, with 16 connectors. Agents act with the signed-in user’s own permissions, so an agent never sees more than the person it works for.
Most organisations run on one or both. We have years of delivery experience across each, so your data foundation and agents fit the tools you have — not the other way round.
Agents and data checks that work in Teams, SharePoint and Outlook, with documents and lists feeding your master data.
Agents that read Drive and Gmail, and data pipelines that land in BigQuery for reporting, with Gemini as one model choice.
Product names belong to their owners. This describes our delivery experience, not a partnership or certification.
Groundfloor-reported results; not guarantees.
A commercial real-estate master data model with 155,000 reference rows, bound to Groundfloor and powering the apps teams use every day.
PremiseHQ sells AI products to commercial real-estate owners across North America. Codemen works as its engineering partner on the products its clients use, including QuadReal, Westfield, Primaris, Colliers, Morguard and Whitman Family Group.
Certificate-of-insurance emails read, matched and investigated automatically, with exceptions raised to the team.
Names and screens are illustrative.
Bring the messiest list you have: vendors, assets or customers. In a 30-minute call an architect will map where it lives and what one clean version would unlock.
Free. No slides unless you ask for them.