Senior AI Architect · CTO, Eir Tec · Sandefjord, Norway

Architecture that survives audits.

I design and operate AI systems for healthcare, finance and the public sector. Twenty years in regulated software, starting in hospital IT. Five products in production today.

Portrait of Håkon Berntsen

Taking new engagements from Q1 2027

In production Eir Tec MediVox Skjld Labs Jaydus.AI InfoDesk

01Selected work

Five products. One backbone.

Each of these runs with real users today: clinicians, patients and operations teams. Pilots are not listed.

CTO, Eir Tec LTD · CEO, Eir Tec AS

Home-based QEEG diagnostics with explainable AI for ADHD and ASD

A four-channel headband records at home. Clinicians receive an explained assessment in minutes instead of weeks, and the clinical journal turns consultations into SOAP notes with on-premise speech recognition. Single-tenant per clinic on Azure UK South, built to UK GDPR and the Data Protection Act 2018.

  • ~7 min time to result, from 7 to 10 weeks
  • +44 % ADHD diagnostic accuracy, internal validation 2025
  • −87 % cost per assessment, versus clinic baseline

Azure UK South · single-tenant · explainable AI · ASR + LLM · NHS DSPT aligned

View at eirtech.co.uk

COO and Systems Architect

MediVox

CE-marked medical documentation AI for Norwegian healthcare. Local language models and on-premise speech recognition turn consultations into structured SOAP notes without patient data leaving the tenant.

Local LLMs · Whisper · SOAP · Norway East

CTO

Skjld Labs

Applied-AI lab for regulated clients. Agents, evaluation pipelines, retrieval systems and multi-tenant SaaS scaffolding, built so the next product starts from a proven base.

Agents · evals · RAG · multi-tenant SaaS

Co-founder

Jaydus.AI

Multi-model platform routing between fourteen-plus language models with shared knowledge bases and SOC 2-aligned controls.

GPT · Claude · Gemini · 14+ models

Founder · since 2002

InfoDesk

IT operations and Microsoft partner. The identity, infrastructure and support foundation under everything above.

Microsoft partner · identity · managed cloud

02How I build

One backbone. Every product.

The same foundation runs under each product I architect, so a new vertical ships in months rather than years, and every one of them passes the same audit.

L1 · Product

What the user touches

Interfaces built for clinicians and operators, with audit views as first-class screens.

Clinician dashboards · Patient apps · Admin consoles
Audit and compliance views · Multi-tenant SaaS shells

L2 · Intelligence

Where the value is generated

Models that can be evaluated, explained and swapped.

LLMs, local and hosted · Speech recognition · Retrieval
Autonomous agents · Explainability · Evaluation suites

L3 · Platform

What it runs on

Single-tenant by default for regulated customers. Data stays in the region it was collected.

Azure UK South and Norway East · Kubernetes · Terraform
PostgreSQL · Redis · WorkOS SSO · CI/CD · Observability

Architecture survives audits

Compliance is designed in on day one, not documented afterwards.

Single-tenant is a feature

Regulated customers get their own deployment. A selling point, not an overhead.

Explainability over magic

A clinical output nobody can explain is a liability.

Reuse the backbone

Identity, speech, models and platform are shared. Products are not.

Open by default

Open source unless patient data or security says otherwise.

Track record

03Ask the architect

Questions about how any of this is built?

The assistant is trained on my work, writing and architecture decisions across Eir Tec, MediVox and Skjld Labs. Ask how a product handles data residency, why a stack was chosen, or how explainability is implemented. It is updated as the work evolves.

How does Eir Tec stay GDPR-compliant?

Each clinic runs in its own tenant on Azure UK South. Patient data never leaves the tenant; retention is anonymised by default, access goes through WorkOS SSO, and every model call is logged for audit.

Which parts of the stack are shared between products?

Identity, speech recognition, model routing, observability and the platform layer. The products on top are not shared.

The assistant answers from Håkon's published work and may be wrong. For anything binding, ask Håkon.

04Get in touch

Bringing AI into a regulated environment?

I work with leadership teams in healthcare, finance and the public sector who need production AI without compromising on data protection. If you have a hard problem, I would like to hear about it.

  • Architecture review
  • Fractional CTO
  • Production delivery
  • Technical due diligence