A Public-Sector Operating System

By Anushka Appala and Dr. Janio Rosales

We are often introduced as an AI company, and we understand why. Artificial intelligence is the part of NaciluzIA that catches the eye — the system that understands a request in Q'eqchi', translates it, drafts the paperwork, and answers by voice. But the description is misleading in a way that matters, and we would like to correct it. NaciluzIA is not, at its core, an AI project. It is a public-sector operating system, and AI is one governed component running inside it.

The distinction is not semantic. It changes what the thing is, what it is for, and how you should judge whether it works.

What an operating system does

Think about what an operating system actually provides. It is not any single application. It is the layer beneath the applications — the part that manages identity, enforces permissions, coordinates the components, and keeps a reliable record of what happened. Applications come and go. The operating system is the durable substrate that lets them run safely and together.

A state needs exactly this and largely lacks it. Today a citizen's identity lives in one silo, a social program in another, a payment system in a third, and none of them share a common layer for who may do what, under which rule, recorded how. The result is the everyday failure we built NaciluzIA to close: a person owed a benefit travels hours to an office, is turned away for a missing form, and tries again next month, while the state already holds every fact it needs.

A public-sector operating system supplies the missing layer. It manages identity. It enforces which official is authorized to sign which action. It coordinates the components — the citizen-facing interface, the AI, the rules engine, the payment instruction. And it keeps the record: a tamper-evident, hash-chained ledger, valid under Guatemala's electronic-signature law (Decreto 47-2008), where every step is written as it happens. The applications built on top can change. The substrate — identity, permissions, rules, record — endures.

AI as a governed component

Placing AI inside this frame is what keeps it in its proper place. In an operating system, no single component gets to override the whole. A well-designed system does not let one process seize the machine, bypass permissions, or corrupt the record. It runs each component with exactly the privileges it needs and no more.

That is precisely how AI runs in NaciluzIA. It is a powerful component with sharply bounded privileges. It may understand, translate, draft, explain, and flag. It may not approve a benefit, suspend a right, or authorize a payment. Those binding actions are reserved for other components: a deterministic, published, versioned rules engine that executes the law, and a named human official who signs. The AI surfaces evidence and patterns for legally authorized human review, and the operating system enforces that boundary the way an OS enforces a permission — not as a promise, but as a property of the architecture. AI recommends, policies decide, officials oversee, audits verify. Each is a component with its own role, and the substrate holds them in relation.

This reframing dissolves the fear that usually accompanies AI in government — that the machine will quietly take over the decision. It cannot, for the same reason an ordinary application cannot rewrite the kernel. It runs where it is permitted to run and nowhere else.

Why the substrate is where the value is

Seen this way, the returns come from the operating system, not from any one clever feature. Deloitte estimates that AI could free up to 30% of government-workforce time within five to seven years (Deloitte, AI-augmented government). But freed time only becomes value if there is a substrate to route it into — if the identity is verified, the rule is published, the signature is authorized, and the record is kept. Absent that layer, automation just produces faster decisions no one can stand behind. The operating system is what converts raw capability into accountable service.

The identity layer is the clearest case. McKinsey estimates that comprehensive digital ID could unlock economic value equivalent to 3 to 13% of GDP by 2030, around 6% in emerging economies (McKinsey, 2019). That value does not come from the ID card. It comes from everything a trustworthy identity layer makes possible on top of it — benefits that reach the right person, payments that resist leakage, services that a citizen can complete in minutes rather than weeks. Identity is an operating-system function, and the GDP is the return on having built the substrate right.

This is why we resist being called an AI company, and why our first step is so deliberately modest: a three-month MVP for one MIDES social program, in one municipality, in Spanish and Q'eqchi', on simulated data. On its face it is a small demonstration — application time cut from weeks to minutes, with 100% of decisions kept human. But what it really proves is that the substrate works: that identity, rules, signature, and record can hold together while an AI does the understanding and drafting inside the boundaries set for it. Get the operating system right in one program, and every program that follows inherits it.

An AI feature impresses for a quarter. An operating system compounds for a generation. We are building the second thing, and the AI is simply one of its best-behaved components.

Cada decisión, a la luz.

— Anushka Appala and Dr. Janio Rosales