Techbanq

AI · government

A government service is judged in Arabic, by everyone

The AI a public body needs is not a novelty bolted onto a portal; it is a citizen-facing service that answers correctly in Arabic first, runs on infrastructure that keeps national data inside the country, and can show why it said what it said when a citizen relies on it. A service that works only for the confident, English-typing minority has not served the public it exists for.

Enterprise delivery credentials

  • ISO

    ISO/IEC 27001

    ISO/IEC 27001 certified

Certifications & registrations

Arabic is the service

Why a public service cannot treat Arabic as a translation

We do not have an office in Saudi Arabia

We serve organisations there from our Riyadh office and our engineering hubs in Lahore and Karachi, travelling for the parts of an engagement that need to happen in person.

A common way a government AI project disappoints is to build it in English and add Arabic at the end. The result understands formal, well-spelled Arabic and struggles with everything else — the dialect a citizen actually speaks, the abbreviations they type on a phone, the sentence that mixes Arabic and English mid-thought. For a private company that gap loses a customer; for a public body it excludes a citizen from a service they are entitled to, which is a different order of failure. Arabic is not a language the service is translated into. It is the language most of the public will address it in, and it has to be first.

Building Arabic-first changes the order of the work: the service is designed and tested against how citizens really write before an English surface is added, not after. It also changes what "correct" means. A public answer that is fluent but subtly wrong about an entitlement is worse than no answer, because a citizen will act on it. So the service is grounded in official policy and built to say when it does not know, rather than to sound confident about a rule it has invented.

From question to resolution

How a citizen's question moves through a service that has to be right

A public service is measured on whether an ordinary citizen gets the correct answer in their own words, and reaches a person when the automated path cannot help. These are the stages that decide whether it does.

01

Asked in real Arabic

A citizen types a question in dialect, with abbreviations, perhaps switching to English partway through.

The service understands the intent behind the informal wording instead of only recognising formal, correctly spelled text.

  • Arabic NLU
  • Web and app
02

Answered from policy

The citizen needs the actual rule that applies to their case, not a general summary.

The answer is drawn from the official policy that governs the service, with a reference, so it can be checked and is not invented.

  • Policy source
  • Grounded retrieval
03

Kept inside the country

The question and any personal detail in it are data the state is required to keep in-country.

The service runs on infrastructure that keeps that data within national borders rather than sending it to an external model.

  • In-country hosting
  • Private model
04

Escalated to a caseworker

The case turns on judgement, an exception or an entitlement the service must not decide alone.

It hands the case to a named caseworker with the context intact, so a person makes the decision the citizen can hold the body to.

  • Case system
  • Human review

The non-negotiables

Four things a public-sector deployment cannot skip

A private company can decide these are optional. A public body serving every citizen cannot, and a deployment that skips one of them fails the people it is accountable to.

01

Arabic that meets the citizen where they write

Dialect, abbreviations and mixed Arabic-English are the normal case, not the exception, and the service is built and tested for them from the start.

02

Data that stays in the country

Citizen data carries residency obligations a public API breaks by sending it abroad, so the service runs where the data is required to stay.

03

An answer that can be checked

Every answer traces to the official policy it came from, so a citizen relying on it and an official reviewing it are looking at the same source.

04

A person the citizen can reach

When a case needs judgement, it reaches a caseworker rather than a dead end, because a public service cannot leave a citizen with nowhere to go.

Our role

What we build and what stays a matter of public accountability

  1. 01

    Arabic-first, then everything else

    We design and test the service against how citizens really write in Arabic before an English surface is added, so the majority is served first rather than last.

  2. 02

    Hosted where the data must live

    We deploy on infrastructure that keeps citizen data in the country, using a private or sovereign model where residency requires it rather than a public API.

  3. 03

    Grounded, with a way to a person

    We ground the answers in your official policy and build the escalation to a caseworker, so the service is accurate on the routine and honest about what it must not decide.

What a citizen is entitled to, and which decisions the state must make through a person, are matters of policy and public accountability that stay with the body; we build the service that answers the routine correctly in Arabic, keeps the data in the country, and routes the rest to a caseworker.

Common questions

What buyers ask

Does government AI mean citizen data leaves the country?+

It does when it is built on a public API, and that is exactly why we do not build it that way for citizen data. Calling an external model sends the question — and any personal detail in it — outside national borders to infrastructure the state does not control, which is the transfer residency rules exist to prevent. The alternative is a service that runs on hosting inside the country, using a private or sovereign model where the sensitivity requires it, so the data stays where it is required to stay. We plan a public-sector deployment around that constraint first, because for citizen data the location of the model is not a detail but the starting decision.

Will it work for citizens who write in dialect, not formal Arabic?+

That is the whole point of building Arabic-first, and it is where projects that add Arabic at the end fail. Citizens do not write like a policy document; they write in dialect, use abbreviations, misspell under a small keyboard, and switch to English partway through a sentence. A service that only understands formal, correctly spelled Arabic quietly excludes most of the public it is meant to serve. We design and test against how people actually write, treating the informal case as the normal one, so an ordinary citizen gets a correct answer without having to phrase their question like an official form.

Can we be held accountable for an answer the system gave a citizen?+

Yes, the body remains accountable, which is why the service is built to make that accountability manageable rather than to hide behind the machine. Answers are grounded in your official policy and carry a reference, so what a citizen was told can be checked against the same source an official would use, and the interaction is logged. Decisions that turn on judgement or an entitlement are not left to the model at all; they are escalated to a caseworker who makes the call. The design assumes you will have to answer for what the service said, and it gives you the record and the human checkpoint to do so — rather than an automated decision no one can explain after the fact.

Start with your data

A scoping session on one citizen service that fixes the Arabic handling it needs, where the data is required to sit, and the point at which a case reaches a human caseworker.

Start the conversation