URIM · The systems an institution cannot buy off a shelf

What we build

Not a product looking for a department.

Four kinds of system, starting with whichever costs you the most. We take the process that is bleeding time or money, build the smallest thing that fixes it, and then the one next to it.

◆ 01

The work behind the form

Most digitalisation stops at intake. The citizen or the supplier now submits online. Behind the counter someone still reads the PDF, re-types it into a second system and forwards it by email. We automate that half: extraction, completeness checks, routing, the first draft of the decision.

What is left for a person is the judgement call. Which is what you hired them for.

◆ 02

Systems that talk to each other

One case touches four departments, three databases and a mailbox. Nobody owns the whole of it, so the weeks disappear in the handovers and the answer to “where does this stand” is a phone call.

We build the layer above: one case, one status, every parallel review visible with its clock running, without replacing the systems underneath.

◆ 03

The decision layer

Under the operations sits the harder question: which project, which supplier, which allocation. We hold the numbers, the constraints and the risks in one model and resolve them into one recommendation, argued.

Every figure traceable to its source, and the counter-case included. The person who signs has to defend it in a room we are not in.

◆ 04

AI that never leaves the building

The reasoning core is ours end to end: a from-scratch engine, not a wrapper around somebody else's API. It runs on hardware you own, air-gapped if that is the requirement, with no call leaving the rack.

Most of what is sold as private AI is an API key and a data-processing agreement. This is neither.

We integrate with what you already run, and we hand over the source. Nothing in the arrangement requires you to keep us.

YOUR INFRASTRUCTURE · YOUR DATA · YOUR KEYS

Where we work

The same method, five very different rooms.

What they share: the decision is expensive, the information is scattered across systems that were never meant to meet, and at the end of it a named person has to sign.

Raw materials & mining

A digital twin of the operation, covering production, cost, ore grades and equipment, so margin is visible rather than argued. Above it, when owning beats importing.

Infrastructure & construction

Proposals turned into decisions that finance: which project holds, with every number backed well enough to survive a board and its lenders.

Logistics

At the Port of Dar es Salaam cargo clears in nineteen days on average. That is working capital sitting in a container. We surface the bottleneck, then design the response.

Agriculture

Yields, prices, weather and supply in one picture: when an investment pays, and which risk to address before it arrives, whether currency, regulation or supply.

Public administration

Permits, registrations, licensing and procurement: the queues a city's economy runs through. Newer ground for us, and the one where better software shows up fastest in the numbers. The city brief →

Proven where the data is thinnest, in Africa and along the trade corridors that run out of it. A system that holds there holds anywhere.

BOARD MEMO · EVERY FIGURE SOURCED · EXPORTED IN ONE CLICK

How we work

One process. Measured before, measured after.

Transformation programmes fail because they promise everything and prove nothing for three years. We go the other way round.

Week zero We sit with the people doing the work and time it. Not the process as documented, but the process as performed, including the workarounds nobody wrote down.

The pick One process that costs real money and has a number attached to it. Not the whole transformation programme.

The build The smallest system that moves that number, on your infrastructure, alongside what you already run. Nothing is ripped out to make room for us.

The proof The same measurement, three months later, published to whoever asked for the budget.

The spread Then the process next to it. Cheaper every time, because the hard part, the integration, is already paid for.

The constraint Every proposal we put in front of you closes on what could stop it. A capacity you do not have, a system we may not get an interface to, a political window that could shut. You hear the risk from us before you hear it from your own team.

We do not unlock one process.
We unlock everything it touches.

Live terminal

Provenance you can click.

Everything above is a claim until you can check one. So here is a working slice of the platform, open to anyone: every figure resolves to a World Bank series held in a local data bank. Select territories, compare them, and follow any number back to where it came from. That is what “traceable” means in practice.

urim · sovereign data terminal

6 indicators · 217 territories · 1960–2025

Load the terminal here, or open it full-screen for the comparison view.

Open full-screen ↗

Data source: the World Bank Open Data Indicators API. Series are pulled on a schedule into a local data bank that the terminal reads from, so nothing is fetched from a third party while you use it. Each indicator carries a link back to the series it came from.

Why it holds

Built to withstand the person who has to sign.

A recommendation is only worth what it survives: the board, the council, the audit, the counter-case. Four properties make ours hold, and none of them can be bought from an external API.

◆ 01

It survives the legal review

Because no third party sits in the path, the questions that stall these projects all have one-word answers. Where does the data go, who else can read it, under whose jurisdiction.

DSGVO becomes a conversation rather than an obstacle, and the security team's objection is answered by the architecture instead of by a clause in a contract.

◆ 02

Provenance by architecture

Citation isn't a formatting step bolted on at the end. Sourcing is carried through the pipeline, so a claim without a traceable origin cannot reach the page, or the decision that quotes it.

◆ 03

The counter-case, included

Every recommendation closes on the assumption that could kill it, whether that is the exchange rate it rests on, the counterparty it trusts, or a legal basis that will not stretch that far. A decision-maker told only the upside cannot act; one shown the downside can.

◆ 04

A compounding record of being right

Every recommendation is timestamped and revisited against what actually happened. The track record is the product, and it only accrues to whoever started first.

The people

Named to you, under NDA, before the first briefing.

Institutions buy from people they can name, so the names, CVs and references travel to you directly rather than sitting on a public page. Three founders; what stands here is what each is for. You deal with them, not with an account layer in front of them.

Strategy & Mandates

Commissioned by a national operator's leadership to build a market-entry model end to end; secured a government mandate for an SDG-aligned procurement platform. The method the system encodes is theirs, executed by hand before it was engineered.

Engineering & Delivery

A former technology operator who ran their own GmbH and builds production systems at the pace of a team five times the size, with the judgement to know which half of a roadmap to refuse.

Structure & Critical Review

Converts a mandate into a plan: scopes the work, governs how the team runs against it, and pressure-tests every assumption before a client has to, asking what breaks while it is still cheap to find out.

The three above are the founders, and the work is theirs. Capacity beyond them is drawn from engineers and specialists we have already delivered with, brought on as a mandate requires and named to you before they start, under the same NDA. Scale is not the constraint on what we can accept.

We build the systems an institution cannot buy off a shelf, then leave them inside your walls.