AI for the public sector, from citizen-facing service intake to procurement, internal knowledge and regulatory monitoring.

Article 49 of Regulation (EU) 2024/1689 requires public authorities deploying high-risk AI listed in Annex III to register the system in the EU public database before putting it into service. The register entry covers the provider, the deployer, the high-risk classification, the intended purpose, and the relevant fundamental-rights impact assessment. We file the entry as part of the build phase and keep it current as the system evolves. The register is public, so the entry is written to be defensible to a journalist as well as to the supervisor.

Procurement is the first conversation, not the last. In the discovery phase we identify the procurement path that fits: a framework agreement the body already holds, a dynamic purchasing system, an innovation partnership under Directive 2014/24/EU Article 31, or a design contest. The build plan is then aligned to that path so the contract and the work plan match. Where the body needs to run a fresh tender, we provide the technical specifications and evaluation criteria as input.

Yes, where the tender admits SMEs and individual specialists. We satisfy the standard self-declaration (ESPD), insurance, conflict-of-interest and exclusion-grounds requirements. For large multi-vendor frameworks we partner with prime contractors and take on the AI-specific lots. We do not bid where the tender requires capacity we cannot deliver to the bar.

Every system ships with first-class support for the official language or languages of the deploying body, including Lithuanian, German, French, Spanish and the other 20 EU official languages. Citation strings, audit logs and the citizen-facing UI are localised and reviewed by a human translator before go-live. Where the body operates across multiple languages, the citation chain points to the language-specific authoritative version of the source.

Every citizen-facing decision the system contributes to carries a cited rationale. The citizen can request the underlying chain - source policy or law, the version of the system that produced the output, the civil servant who signed any non-trivial step. The interface for that disclosure is part of the build, not a manual back-office process.

Article 22 prohibits decisions producing legal or similarly significant effects from being made solely on automated processing. In the public sector the safe path is human-in-the-loop on every decision that affects a citizen's rights or benefits. We build the workflow so the AI structures the decision packet (facts, applicable rules, comparable cases, draft rationale) and the responsible civil servant signs the actual decision with the ability to override. Every override is logged with reason codes.

By default, all processing and storage runs in EU regions on infrastructure under EU jurisdiction. We support member-state pinning when a regulator or contract requires it (Germany-only, France-only, Spain-only, Lithuania-only). Citizen data lands in immutable EU object storage with hashes recorded in the audit log. We do not train any model on citizen data, full stop.

A first engagement targets one workflow with a measurable baseline, runs 4 to 12 weeks to production, and lands as a single signed-off system inside one source-of-truth platform. Common first scopes are: citizen-intake assistant for one service line, procurement document automation on one category of tenders, internal-knowledge AI for one ministry's legislation database, or regulatory monitoring for one directorate. Submit a project with the workflow you have in mind and the procurement path you can use, and we scope the discovery phase before any code is written.