Deployment is where AI in credit works or stalls.
- A credit lead who knows how your committee decides.
- An engineer who builds it into your systems.
- Your people who run it once it is live.
A team on your floor
A credit lead and an engineer work inside your credit operation, beside your people, in Dubai, Abu Dhabi and Doha. Not a vendor on a call.
Credit lead
Makes it count.
Picks the job that pays first, writes down how your people decide, and measures what changed.
Engineer
Makes it work.
Builds the workflows into your systems. Solves access, data and controls on site, while it is being built.
It goes in from your floor, and goes live only when it passes your hardest cases.
Phase 1
Pilot
You give us 10 to 20 files you have already decided. We write your memo template and credit policy into the system and run them, and you read our memo beside the one your committee signed. It takes four to six weeks, and it is free.
→
Your memo, every figure cited to its page
→
In our sandbox, or in your own cloud account
→
Those files become the test it must pass
How it works
What reads what, what is code, and what only a person does.
1Received
Code, no model
The pack is opened in code: a .zip is unpacked, and each file's pages are counted.
2Triage
Model: Claude Haiku 4.5
Rules read every file name and folder first, then Claude Haiku 4.5 looks at the first page of each document, as an image for a scan or a photo, as text for the rest, and names it: its type, its period, whose it is and its language.
3Extraction
Model: Claude Sonnet 5.5
Every figure comes back with its document and page, or it is left out. Nothing is guessed.
4Checks
Model: Claude Sonnet 5.5Code, no modelA person
Each check names both of its sources, and a check the pack cannot run is left out.
5Your confirmation
A person
Nothing reaches the score until a person confirms the figures against the page and corrects any misread.
Your hardest past files
Borrower to decision, replayed
The rule is written down. It runs again.
4 of 5 cases pass
When every file matches
Phase 2
Fit and go live
We connect your systems, set up your memo, documents and signers, install it in your own cloud and train your team. The first stage is live in about two months, and only once it gets your hardest past files right.
→
Your systems connected, in your cloud
→
Your memo, documents and signers set up
→
Live only when your hardest files pass
Phase 3
Run and extend
We run it. We catch the failures, fix them and extend your system as the work changes. The code, the workflows, the credentials and the exception paths stay yours.
→
Watched, with exceptions routed to a person
→
Extended as the work changes
→
New formats caught the first time they arrive
Borrower to decision
LiveActivity
- Policy adds a sector limit
Waits on the head of creditApproved by the head of creditWaitingApprovedChange, today
- Spreading reads a second audit formatReads a second audit format
Change, 9 Sep
- Goes liveEvery past file matched.
Live, 18 Aug
Built by us. Yours from the first day: the code, the workflows and the credentials sit in your accounts.
Deploy where you run
Matched to your infrastructure and to what your regulator expects.
Where it runs
The models that read your files, and where they run, are yours to choose.
Deployment
One workspace, wherever you run it.
How it works
What reads what, what is code, and what only a person does. Read from the instance's own model routes.
1Received
Code, no model
The pack is opened in code, and each file's pages are counted.
2Triage
Model: Claude Haiku 4.5
Each document is named from its first page: its type, its period, whose it is.
3Extraction
Model: Claude Sonnet 5.5
Every figure comes back with its document and page, or it is left out.
4Checks
Code, no model
Figures from two documents are set side by side, each check naming both sources.
5Your confirmation
A person
Nothing reaches the score until a person confirms the figures.