Service · Software development

Custom AI Act compliance software development.

The EU AI Act (Regulation 2024/1689) requires organisations that use AI to maintain a demonstrable register, risk classification, technical documentation and human oversight. We build the platform in which you manage all of this centrally, tailored to your AI portfolio, your governance and your auditors.

AI system registerRisk classificationHuman-in-the-loopAudit trail

Why a custom AI Act platform rather than a generic GRC tool.

The AI Act (Regulation 2024/1689) has been in force since August 2024 and is being phased in until August 2027. It is the world's first comprehensive AI regulation and uses a risk-based model: unacceptable risk is prohibited (government social scoring, real-time biometric identification in public spaces, emotion recognition in the workplace), high risk is heavily regulated, limited risk comes with transparency obligations, and minimal risk faces hardly any requirements. High-risk AI carries substantial obligations: an AI system register, a risk management system covering the entire lifecycle, data governance with bias checks, technical documentation, logging, transparency towards users, human oversight, cybersecurity requirements, CE marking and registration in the EU database.

Generic GRC and privacy tools (OneTrust, Holistic AI, comparable suites) are often good at the legal layer, but they rarely connect to the ML process itself: how models are trained, which datasets go into them, which metrics flag drift, and how a human-in-the-loop fits into the production flow. We build the piece that sits between your AI engineers, your compliance officer and your board, either as the core of your AI governance or as a specialised layer alongside your existing tooling. For the broader strategy, this ties in with our approach to enterprise AI.

If you bring the AI system to the European market as a product, the requirements of the Cyber Resilience Act apply as well: reporting deadlines, security updates and a component overview. We set up the documentation so that you maintain it once and can use it in both places.

Three flavours of AI Act software.

Depending on where you stand: just starting to take inventory, in the middle of classifying your first high-risk system, or working on a full governance platform for your entire AI portfolio. We will advise which variant fits in the first working session.

Compact project · fixed sprint budget

AI system register

A central register of all AI systems in your organisation: purpose, dataset, model version, supplier, owner, risk category and compliance status. It is the starting point for any AI Act implementation and the thing many organisations search for, rightly so, because without a register you do not know what falls under the law.

System inventoryOwnershipVersion controlAudit export
Mid-sized project · fixed sprint budget

Compliance dashboard with risk flow

On top of the register: a workable risk assessment flow from project start to decommissioning, classification into unacceptable/high/limited/minimal, and a dashboard that gives the board and compliance officer an at-a-glance view of the status per system. Including a technical file vault and model cards.

Risk classificationLifecycle managementDocumentation vaultStatus dashboard
Larger project · fixed sprint budget

Full AI governance platform

A mission-critical platform with bias monitoring, drift detection, human-in-the-loop workflows, incident management, vendor management for third-party AI, and integrations with your ML stack (MLflow, SageMaker, Vertex AI, Azure ML). Suited to organisations with dozens of AI systems, regulated sectors or a clear board-level governance need.

Bias & fairnessDrift detectionVendor riskML stack integrations

What you get at the end.

A production-ready AI Act platform, plus everything around it to enable your compliance team, AI engineers and auditors to work independently.

  • AI system register + classification flowCentral register with risk categorisation under the AI Act, ownership per system and automatic trigger events for reclassification when models change.
  • Documentation vaultVersion-controlled storage for technical files, model cards, data sheets, DPIAs, FRIAs and risk assessments, delivered in the format regulators expect.
  • Logging and audit trailDemonstrable logging of model runs, decisions, human reviews and changes to training data, in line with the AI Act's traceability requirements.
  • Bias and drift monitoringFairness metrics per protected group, drift detection at feature and outcome level, and alerts for your data science team as soon as a model moves outside its tolerance range.
  • Human-in-the-loop workflowA configurable review and sign-off flow for decisions made by high-risk systems, with SLAs, escalation and a complete audit log.
  • Incident and vendor managementRegistration of serious incidents in line with the AI Act, supplier risk assessment for third-party AI, and post-market monitoring per system.
  • Codebase, training and maintenanceFull source code, a deployment runbook, two training sessions for compliance and data science teams, and an optional maintenance contract for further development.

When AI Act software is the right investment.

Four situations in which we support organisations. If you recognise one of these patterns, we'd be happy to discuss the approach further.

High-risk

You use AI in a high-risk application

Recruitment, credit scoring, education, critical infrastructure, law enforcement, biometrics, migration or justice. The AI Act imposes its heaviest obligations here: a register, technical documentation, human oversight, CE marking and EU registration.

Scale

Your AI portfolio is outgrowing spreadsheets

You can still manage one or two AI systems in Excel. At ten or more, version control, data governance and classification become unmanageable without a central system that connects to your ML stack.

Oversight

You fall under a sector regulator

DNB, AFM, AP, ACM or the Health and Youth Care Inspectorate: they will assess AI systems in conjunction with sector-specific rules. A demonstrable register and risk management system is then practically a requirement to get through an audit.

Governance

Your board wants control over AI risks

Board members are held accountable for AI-driven decisions. A dashboard showing the current compliance status per system lets you fulfil that responsibility without relying on verbal updates from the data science department.

Which modules we typically build.

AI system register. The heart of your platform. For each AI system, we record its purpose, owner, supplier, datasets, model version, deployment location, business impact and risk category. Fields are dynamic depending on the category: a high-risk system requires considerably more documentation than a minimal-risk chatbot. The register is searchable, exportable in audit formats and can be connected to your CMDB or MLflow instance.

Risk assessment flow. A structured wizard that guides a new AI project from idea to deployment. It asks the right questions at each stage, classifies automatically as unacceptable/high/limited/minimal, and includes a gate mechanism that prevents a system from going live without completed documentation. For high-risk systems, this includes a Fundamental Rights Impact Assessment (FRIA) under Article 27.

Documentation vault. Version-controlled storage for technical files (Article 11), data sheets, model cards, training data overviews, evaluation reports and declarations of conformity. Templates for the standard formats regulators expect, with version history and a sign-off flow per document.

Human-in-the-loop module. Configurable review workflows for decisions made by high-risk systems. For each use case, you decide which decisions can be automated, which require human sign-off and which are escalated. Includes SLA tracking, queue management and a complete decision audit log for later reconstruction.

Monitoring layer. Drift detection on input features and output distributions, fairness metrics per protected group, accuracy tracking against benchmarks, and alerts for your data science team. We integrate with MLflow, SageMaker Model Monitor, Vertex AI Model Monitoring or a custom pipeline, depending on your stack.

Incident and post-market monitoring. Logging of serious incidents in line with the reporting obligation under the AI Act, root cause analysis, corrective actions and reporting flows to your national supervisory authority. For ongoing post-market monitoring per system, we build the part that can be automated; the legal part stays with your compliance team.

Vendor management. For third-party AI systems, such as a recruitment tool you have bought that includes an AI component, or a SaaS supplier that uses LLMs, we record the vendor, their declarations, their classification and the risks to your organisation. When a contract is renewed, the platform triggers a reassessment.

Who we work with.

The AI Act affects every organisation that uses AI in the EU, but the impact varies widely by sector. Four groups we often work with.

Financial sector

Banks, insurers, asset managers

AI used in credit assessment, fraud detection and life insurance pricing almost always counts as high-risk. It also overlaps with DORA (operational resilience) and sector-specific requirements from DNB and the AFM. We link AI Act compliance to your existing risk management, without duplicating administration.

HR & recruitment

Organisations that use AI in selection

CV screening, video interview analysis, skills matching: these applications are explicitly listed in Annex III as high-risk. This applies both to your own recruitment tooling and to your suppliers' tools. A register combined with vendor management is then practically indispensable.

Public sector & healthcare

Government, education, healthcare, critical infrastructure

Almost every AI application in the public sector touches high-risk domains: educational assessments, justice, migration, social benefits, critical infrastructure and, in healthcare, the overlap with the MDR. On top of that, transparency requirements for automated decision-making towards citizens often apply.

Tech & scale-ups

Companies that sell AI as a core product

If you supply AI functionality to other companies, you are often the provider under the AI Act and carry the heaviest obligations. For SaaS companies selling to enterprise customers, a demonstrable register and technical file are now becoming a sales requirement, as your customers will ask for them during their own audits.

How an AI Act project works.

1

Introduction and AI inventory

Together we map out which AI systems you currently have or are building, which suppliers are involved, which processes depend on them, and where the highest risks lie. For organisations without an existing register, this is often where the real surprises come to light.

2

Workshop: scope, classification and governance

A working session with your compliance officer, data science lead and business owners. We map your AI systems to the AI Act categories, link them to GDPR overlap, and decide which modules are in scope for the first release.

3

Building in sprints

A working build every two weeks. First the register and classification flow, then the documentation vault and audit log, followed by monitoring and human-in-the-loop. Your team tests along the way, and your compliance officer gives feedback on the legal conformity of each module.

4

Rollout, training and post-market monitoring

A phased rollout per business unit, training for compliance and engineering teams, and ongoing maintenance for security patches, regulatory updates and extensions. The AI Act will keep evolving over the coming years, and your platform needs to evolve with it.

Frequently asked questions.

What organisations usually want to know before we start an AI Act implementation.

When does our AI system have to comply with the AI Act?
The Act has been in force since 1 August 2024. The prohibitions on unacceptable AI have applied since February 2025. GPAI obligations for model providers have been active since August 2025. The stricter rules for high-risk AI are being phased in until August 2027. For most organisations, that means starting now with taking inventory and classifying your systems, rather than waiting until 2027.
How do I know whether my AI system is high-risk?
The AI Act explicitly lists the high-risk domains: critical infrastructure, education and recruitment, creditworthiness and life insurance, law enforcement, migration and border control, the administration of justice, and certain biometric applications. A risk assessment in our classification flow translates your specific use case into the right category and determines which obligations apply.
What is the difference between GPAI obligations and high-risk obligations?
GPAI (General Purpose AI) obligations apply to providers of foundation models, such as OpenAI, Anthropic and Mistral. They must supply technical documentation for the model, respect copyright and report on energy efficiency; for systemic-risk models, they must also carry out model evaluation, red-teaming and serious-incident reporting. High-risk obligations apply to whoever deploys such a model in a regulated domain such as critical infrastructure, recruitment, credit assessment or law enforcement. You can perfectly well be both a user of a GPAI model and a provider of a high-risk AI application, in which case you take on both roles, and your platform needs to keep that distinction clear.
What obligations apply to limited-risk AI such as chatbots?
Limited-risk systems, such as chatbots, deepfakes and AI-generated text and images, are mainly subject to transparency requirements. Users must know they are talking to an AI, content must be labelled as AI-generated, and deepfakes must be marked as such. There is no registration obligation and no technical file, but there must be clear disclosure. Our platform also records this category, for example for the central list for your board of directors, but with lighter documentation requirements than for high-risk systems.
Does this replace OneTrust, Holistic AI or similar suites?
Not necessarily. For enterprise organisations with an existing GRC stack, we are more likely to build a specialised AI layer that works alongside it: the ML side (data governance, drift, fairness, model versions) is handled by us, while the legal side (policy management, awareness) can stay in your existing tool. For organisations without a heavy GRC investment, our platform can become the central hub.
How does the AI Act relate to the GDPR?
There is a lot of overlap, but one does not replace the other. The GDPR covers personal data, while the AI Act covers AI systems, and the two intersect when your AI processes personal data. In practice, this means a DPIA is often needed in addition to an AI risk assessment, and an automated decision under Article 22 of the GDPR will almost always also fall under the high-risk rules of the AI Act. Our platform links both processes.
What determines the lead time and costs?
Mainly: the number of AI systems that need to go into the register, the maturity of your existing ML stack, how many data sources are connected for monitoring, and whether you are building a specialised layer or a full governance platform. A register for a single department can be up and running in a few sprints; an organisation-wide governance platform is a project spanning several sprints. After the inventory, we provide a concrete plan.
Do you work with our legal team and data science department?
Almost always. The AI Act sits at the intersection of case law and technology, so we don't build a page your lawyers can't follow, or technology your data science department can't connect to. That means workshops during scope definition, peer reviews of the classification flow, and knowledge transfer in the final sprint so your team can manage the platform independently.

Talk to us about your AI Act compliance.

A free, no-obligation half-hour introduction. We listen to your AI portfolio, your governance questions and your deadlines, and give you direction on a platform that fits. Related topics: DORA compliance software, KYC/AML compliance software and custom risk management software. For automation around AI decision-making, it ties in with AI agents.

Edit content