Auditable by design
Every transaction leaves a trace and every automated decision can be reconstructed. In the public sector this is not one more feature: it is the requirement that decides whether the system may operate at all.
Two systems for the Panamanian state, both in production: the Electoral Tribunal multiservice kiosks, with 100 units delivered, and the platform behind the AstraZeneca vaccination rollout in Panama. This page is what I can build; the rest of the site is the evidence that I already have.
Applied AI aimed at concrete institutional problems — and the full information system underneath it. One without the other never reaches production: a model with no system is a demo, and a system without the right data is an expensive filing cabinet.
Round-the-clock service over WhatsApp or the web, with graduated autonomy levels and an immediate off switch. The agent answers with the institution’s real information, escalates to a person when it should, and leaves a trace of every conversation.
Retrieval over the entity’s documents and current regulation, with the source cited in every answer and an explicit path to saying "I do not know". The expensive failure mode of these systems is answering confidently with no grounding; this is designed so that never happens silently.
Reading, classifying and extracting data from documents somebody types by hand today: OCR over scans, validation against the entity’s systems, and human review on the doubtful cases. The point is not to remove the officer from the loop: it is that they only look at what genuinely needs judgement.
Transfer learning and models trained on the institution’s own data, running inside its own infrastructure. It is the way out when the data cannot leave the country, when the domain is too specific for a general model, or when the per-call cost does not add up at the scale of a public service.
With the real operational history you can anticipate load by site, day and procedure, and staff for it before the queue forms rather than after. The same basis serves to prioritise case files by urgency or by risk.
Identification and identity verification with an in-house pipeline: detection, embeddings, anti-spoofing and edge deployment. It runs on low-cost hardware and integrates with terminals the institution already has installed.
The system that holds a complete procedure: application, requirements, states, deadlines, owners, notifications and resolution. With per-role profiles and permissions, an audit trail on every change, and the file reconstructable at any point.
Rates, surcharges, subsidies, fines or settlements where each case is computed under the rule in force on the day that applies, even when a period straddles a change in law. It is what saves recalculating half a register by hand every time a table changes — and what makes the result defensible under audit.
No institution starts from zero. Connecting to legacy systems, the databases nobody may touch and the third-party hardware already bought, without demanding everything be replaced first. Modernisation that requires throwing out the old never actually happens.
Ingestion, processing and aggregation of the real operation: service times, volumes by site and by procedure, historical series and dashboards. It is what turns a deployment into a budget argument and a suspicion into a decision.
Self-service units issuing documents, taking card payment on the kiosk itself, running continuously, with remote management of the whole fleet: status, updates and telemetry without sending anyone out to the site.
Tickets, scheduling, counter calling and occupancy control, across several sites and in real time. With calling displays and the citizen’s full path inside the building, not just the moment they book.
Mobile and desktop for people working away from the office: capture with GPS and geofences, offline operation with later synchronisation and conflict resolution. What is recorded in the field lands in the same case file, with no double entry.
Every transaction leaves a trace and every automated decision can be reconstructed. In the public sector this is not one more feature: it is the requirement that decides whether the system may operate at all.
A proven hybrid architecture: data that cannot leave the institution stays inside, without giving up the operational advantages of the cloud for everything else.
The whole stack is open and standard. The institution keeps its code, its data and the ability to change supplier without rewriting the system.
We start with a bounded scope and metrics agreed in advance. If the pilot does not move the numbers there is no reason to scale it — and that has to be visible in weeks, not years.
Do you have a concrete problem in citizen service, identity or automation? Tell me about it and I will say plainly whether I can solve it and what it would take.
hola@giron.aiI answer, not a form.