Intelligence,
made human.
We design and engineer AI systems that regulated industries can actually deploy — with full reasoning transparency, governed data access, and craft at every layer. All running inside your own infrastructure.
Three layers.
One sovereign runtime.
Every component works independently. Together they form the only compliance-assured agent stack built for regulated industries.
Marino Evident
Reasoning transparencyEvery inference step logged, structured and visualised in real time. Full chain-of-thought audit trail for every agent decision — before any action is taken.
Marino Kestrel
Agent credential managementShort-lived, scoped AWS credentials issued per agent task. Every data access governed, approved and auditable. No standing permissions. Ever.
Sovereign LLM
Self-hosted inferenceRuns entirely within your VPC. No data leaves your infrastructure. EU-hosted. Compatible with Mistral, Llama, Qwen — or your own fine-tuned domain model.
Built for decisions
that matter.
Reference implementations across banking, insurance and healthcare — each running on the same compliant agent runtime.
Commercial lending decision support
Agent analyses financials, runs CRO checks, calculates DSCR and leverage ratios, and produces a structured credit memo — with full reasoning trace and credential audit log. Human credit committee decides.
KYC & AML screening
Continuous monitoring agents that screen transactions and onboarding events against sanctions lists and behavioural patterns. Every decision auditable. Every data access governed by Kestrel.
Underwriting assessment
Risk scoring agents that evaluate policy applications against actuarial models and claims history — with explainable outputs that satisfy Solvency II and EU AI Act high-risk AI requirements.
Clinical decision support
Diagnostic support agents running on air-gapped hospital infrastructure. Patient data never leaves the trust boundary. Reasoning traces provide the explainability trail required by the EU AI Act for Class IIb medical AI.
A model that knows
your domain.
We fine-tune open-source LLMs on your institutional knowledge — trained in your AWS account, owned by you, never shared.
Fine-tuned Mistral or Llama hosted on Marino’s AWS eu-west-1 infrastructure. Dedicated inference endpoint. Data sovereign. Fastest path to production.
Fine-tuning runs inside your AWS account on SageMaker. Model artefact stays in your S3. Training data never leaves your infrastructure. Full IP ownership.
For environments with no cloud connectivity. Quantised models optimised for on-premise GPU infrastructure. Healthcare and defence use cases.
Regulation built in,
not bolted on.
Every platform component maps to a specific regulatory obligation. Your compliance team will know exactly what to point the auditor at.
Evident provides a full reasoning trace for every agent decision. Explainability is automatic, not retrofitted.
Kestrel approval gates enforce human-in-the-loop before sensitive operations. Configurable per policy.
Non-repudiable audit trail of all agent credential events. Continuous export to your SIEM.
Real-time anomaly detection on credential usage. Auto-suspension on policy violations.
No standing credentials. All access short-lived, scoped to minimum privilege, tied to named agent identity.
Principle of least privilege enforced at issuance. Every credential scoped to a single task session.
Thinking out loud.
Born from
Marino Software.
Built for the AI era.
marino.ai is the dedicated AI practice of Marino Software, Ireland's leading strategic design and development consultancy. With over 15 years of building software that people love, we bring the same craft-first philosophy to artificial intelligence.
We believe AI should amplify human capability, not replace it. Our teams combine deep machine learning expertise with world-class UX design to build systems that are powerful, trustworthy, and genuinely useful.
From sovereign AI infrastructure to production ML systems, we partner with regulated organisations who need AI they can actually deploy — with full reasoning transparency, governed data access, and compliance built in from day one.
Ready to build
something intelligent?
Whether you're exploring sovereign AI for the first time or scaling existing systems in a regulated environment, we'd love to hear what you're working on.