You do, from the first commit. The repository is created in your account or handed over whole at the end, with the infrastructure definition and the evaluation set.

Which AI model will it use?

Whichever the evaluation set says is best for the job at the cost you can live with. That can be a hosted model, an open model running inside your network, or no language model at all where a classical approach scores higher.

What happens after the thirty days?

The platform is yours to run. Most clients keep it on the hosting plan, which covers it the same way it covers a site; some take it in-house with the runbooks.

Services (08)

Custom SaaS & AI Platforms

A product taken from a question to a running platform, scoped to the complexity you actually need.

⃁ by complexity, not by the hour, quoted in writing

Start a conversation (Or on WhatsApp)

Not sure yet? Check your site first, free.

Typical build
Demos
Fixes after launch

(01)

Priced by complexity, not by the hour

Three tiers, one written scope. A tool does one job: a retrieval assistant over your documents, an integration between two systems, an internal dashboard. A product has accounts, its own data model, an admin and one AI feature. A platform serves several organisations at once, bills them, carries an evaluation harness for its models and runs inside your own cloud account.

What moves a build up a tier is not the number of screens; it is accounts, tenancy, billing and anything that has to be provably correct. We tell you which tier you are in before we start, and the number does not move unless the scope does.

What each tier carries

	Tool⃁ on request	Product⃁ on request	Platform⃁ on request
One workflow or integration
Accounts and roles
Data model and admin
An AI feature
Evaluation harness
Billing
Multi-tenant
Deployed in your cloud

(02)

Ten weeks, in the open

A typical build is ten weeks: one to discover and write the scope, two to prototype the risky part first, six to build, one to ship. Every Thursday, the last working day of the Saudi week, you see running software rather than a slide about it.

The prototype weeks exist because the hardest question in most AI builds is whether the idea survives contact with your real data. We answer that in week three, while changing course is still cheap.

The ten weeks — phases across, a demo every Thursday

(03)

What we build it from

An interface in the browser, an API behind it, a model layer that is a language model where one is warranted and a plain classifier where it is not, a database, and a cloud account that is yours. Proven tools over novel ones: the platform should be staffable by people you can hire in Riyadh next year.

The source, the infrastructure definition and the runbooks are handed over at the end. You can keep us on, or not, and the platform will not notice.

The layers of a platform, and what each one is for

Interface
API
Model layer
Data
Your cloud

(04)

An AI feature is measured, or it is a demo

Before an AI component is built, an evaluation set is drawn from your cases: real questions, real documents, real edge cases, with the answers your people would accept. The target is agreed as a number. The feature is then built against that set, and each week the number is reported beside the demo.

This is how you know it works before launch rather than after, and how you will know, a year on, whether a model update made it better or worse.

A feature climbing its evaluation set, week by week — the accent row is the agreed target

Week 1
Week 10

(05)

Built in Riyadh, deployed in your cloud

The team works from Riyadh, on the Saudi week and in Riyadh time, so the people who scoped the platform are the people you call about it. Interfaces are Arabic-first when your users are: right-to-left is designed in from the first screen, not bolted on at the end, the way this site was built.

Data stays where you need it to. Where the Personal Data Protection Law or a regulator requires it, the platform and its models run inside a Saudi region of your own cloud, and we say so in the scope.

Nothing ships without — the five gates every release passes

Code review
Automated tests
Evaluation set
Security scan
Your sign-off

Gates
Skipped

Questions we get asked

Who owns the code?: You do, from the first commit. The repository is created in your account or handed over whole at the end, with the infrastructure definition and the evaluation set.
Which AI model will it use?: Whichever the evaluation set says is best for the job at the cost you can live with. That can be a hosted model, an open model running inside your network, or no language model at all where a classical approach scores higher.
What happens after the thirty days?: The platform is yours to run. Most clients keep it on the hosting plan, which covers it the same way it covers a site; some take it in-house with the runbooks.

Riyadh, Saudi Arabia

The coffee is on us.

Tell us about the site or the idea, in Arabic or English, and we will come back with questions, then a number in writing. No deck, no pressure, and the first conversation costs nothing.