// ABOUT
Senior AI Architect · Builder · Open-source advocate
Norwegian engineer from Sandefjord. Twenty years building software in regulated environments — and working with AI and machine learning long before OpenAI was a household name.
[ 02 ] // WHO I AM
I'm Håkon Berntsen — born in Oslo, based today in Sandefjord on Norway's south coast, with a chapter in Hammerfest in between. I grew up taking things apart — radios, computers, networks, anything with wires — and I've been doing some version of that ever since. Through every shift the industry has made: dial-up to broadband, on-prem to cloud, scripts to microservices, rules-based automation to large language models. The technologies keep changing; the curiosity is the same.
My career properly began inside Norwegian healthcare. As an administrator for DIPS at Helsenett Øst, I worked on the systems hospital staff used every day — patient journals, lab integrations, secure messaging. That work taught me, early and permanently, that software in regulated environments is not the same craft as software anywhere else: the audit trails matter, the identity model matters, the failure modes matter, and "move fast and break things" stops being a virtue when the thing you might break is patient care.
Since then I have started, run or co-founded more than eight companies — across payments, cybersecurity, media, IT services, healthcare AI, applied AI research and clinical neurotechnology. They look different on paper, but they share one thread: each has been an excuse to push the same architectural ideas a little further, against a slightly harder problem.
One of those chapters took me far north to Hammerfest, where I ran the local newspaper Hammerfestingen as Managing Director until it was acquired by Amedia. It was an unusual detour for an engineer, but it taught me what it actually feels like to run a real business at the other end of the country: payroll, deadlines, print runs, sales, and the quiet pressure of being the person who has to sign certain things. Those lessons stuck.
Outside the companies I'm a husband and a father, an open-source advocate (currently Chairman of Open Info), and a permanent student of how AI actually changes what software can do for people. When I'm not in code or in meetings I'm usually reading research papers, building side projects with my own AI agents, or arguing — politely — about what "explainable" really means in a clinical context.
// BASED IN
Sandefjord, Norway
// LANGUAGES
Norwegian · English
// COMPANIES FOUNDED
8+
[ 03 ] // AI PIONEER
I started building with artificial intelligence and machine learning in 2005. At the time, that meant Bayesian classifiers, decision trees, support vector machines, hand-tuned NLP pipelines and the kind of small neural networks that ran happily on a laptop. There was no API to call, no model garden to pick from, no weekly product launch — and very little hype. You read the papers, you implemented them yourself, and you compared notes with researchers in international forums.
That's where I came up. Long before the modern LLM era — long before GPT-3, long before ChatGPT, and long before "AI" became a household category — I was experimenting with applied machine learning for natural language, document understanding and automation problems that companies had been throwing humans at for decades. I participated in the early international ML communities and have been quietly contributing — and watching — ever since.
Twenty years later the field has changed beyond recognition: GPT, Claude, Llama, Mistral, multi-modal models, autonomous agents, retrieval pipelines, real-time speech recognition. But the fundamentals haven't moved at all. Knowing how the field actually got here — the dead ends, the breakthroughs, the recurring lessons — is what makes today's stack feel inevitable rather than novel. It's also why I trust myself to architect production AI in places where the wrong call has real consequences for real people.
// FIRST AI WORK
2005
7 years before the deep-learning breakthrough
// PRE-CHATGPT
17 years
building ML before mainstream LLMs
// IN PRODUCTION
5+ products
live AI systems I architect today
[ 04 ] // HOW I THINK
After two decades and eight-plus companies, these are the rules I have stopped second-guessing.
// 01
In healthcare, finance and the public sector the architecture is not just a development concern — it is the compliance posture. I design for the audit on day one, not retrofitted later when someone asks for a SOC 2 letter.
// 02
For regulated customers, single-tenant deployment is not bloated overhead — it is the difference between yes and no. I treat tenancy as part of the product, not an afterthought tacked on at procurement time.
// 03
A clinical AI output that no one can explain is a liability, not a feature. Every high-stakes decision in my systems is traceable, attributable and auditable — including the prompt, the model and the data it saw.
// 04
The LLM, ASR, identity, observability and platform layers are the same across every product I ship. Each new vertical inherits a battle-tested substrate instead of reinventing one in a new codebase.
// 05
Open source has shaped my career and I contribute back where I can. Closed code is reserved for the parts that genuinely need it — patient data, customer-specific tuning, security-sensitive logic — never as a default.
[ 05 ] // THE PATH
The starting point was DIPS at Helsenett Øst — Norwegian hospital systems, real clinical data, real consequences for getting the architecture wrong. That formed the discipline: regulated environments, audit trails, single-tenant deployments, careful identity, clean separation of duties. Everything I have done since has been in some way an attempt to apply that discipline at greater speed and bigger scope.
From there I founded InfoDesk AS in 2002 — a full-service IT operation that became (and still is) the operational base under everything else: identity, infrastructure, support and managed cloud. Around it I built and exited a series of companies across cybersecurity (ZecureCode AS), payments, and applied AI advisory (ReadySOFT AS). One of those chapters took me to Norway's far north as Managing Director of Hammerfestingen, the Hammerfest local newspaper, until it was acquired by Amedia.
The current chapter is the AI-native one. MediVox AS brought AI for medical documentation into the Norwegian healthcare market. Skjld Labs AS is the dedicated applied-AI dev shop sitting under the same group. Jaydus.AI is the multi-LLM enterprise platform — GPT-5, Claude 4, Gemini 2.5, fourteen-plus models in one workspace, SOC 2-aligned. And Eir Tec takes the same backbone into clinical neurotechnology — home-based QEEG and explainable AI for mental-health diagnostics.
None of these are unrelated. The operating model is the same in each: pick the patterns that survive contact with users, regulators and scale, ship them as a product, learn what breaks, fold the lessons back into the backbone, repeat. I am, first and foremost, the architect — the companies are how the architecture reaches users.
[ 06 ] // THESIS
Every system I ship sits on the same recurring stack: cloud-native on Microsoft Azure, single-tenant deployments where regulation demands it, large language models composed with automatic speech recognition and autonomous agents, explainable AI for any clinical or high-stakes decision, identity through WorkOS and OAuth, and privacy-first design against UK GDPR, EU GDPR and SOC 2.
Mental-health diagnostics at Eir Tec, medical documentation at MediVox, applied AI research at Skjld Labs and the multi-LLM enterprise platform at Jaydus.AI are not separate projects — they are the same architecture pointed at different problems. New vertical, same backbone.
I am, first and foremost, the architect — the companies are how the architecture reaches users.
[ 07 ] // CURRENT FOCUS
I am CTO of Eir Tec LTD (UK) and CEO of Eir Tec AS (NO) — the same group, deliberately dual-jurisdiction so the platform can serve UK NHS pathways and Norwegian / EU healthcare systems in parallel.
The product: a home-based four-channel QEEG headband paired with a clinician dashboard powered by Explainable AI. Results in ~7 minutes instead of the traditional 7–10 weeks, a measured 44% diagnostic accuracy improvement for ADHD and an 87% cost reduction. Live clinical transcription drives AI-generated SOAP notes, referral letters and clinical summaries.
The platform runs cloud-native on Azure UK South in single-tenant deployments, fully compliant with UK GDPR and the Data Protection Act 2018, with enterprise SSO via WorkOS.
Co-founders: Dr Monica Berntsen — CSO, HCPC-registered psychologist, PhD in neuroscience (ADHD / ASD specialism); Ignacio Zuniga — CEO, ethical-AI and investment-banking background.
[ 08 ] // ROLES
Concurrent leadership across regulated AI, applied research and enterprise platforms.
CTO
Mental-health diagnostics platform; 4-channel QEEG + Explainable AI.
CEO & Co-founder
Norwegian / EU jurisdiction for the Eir Tec clinical platform.
COO & Systems Architect
AI for medical documentation; part of the Skjld AS group.
CTO
Applied AI research and development across regulated verticals.
Co-founder
Multi-LLM enterprise platform — GPT-5, Claude 4, Gemini 2.5, 14+ models, SOC 2.
CEO & Founder
Microsoft Partner since 2002 — enterprise IT and software delivery.
Chairman
Non-profit advocacy for open source and open standards.
[ 09 ] // AI AGENTS
Autonomous agents are how I scale architectural decisions — the same patterns ship across every product. I work daily with OpenClaw, LangChain and a growing set of custom agent frameworks tuned for clinical, regulated and enterprise contexts. My personal agent, Dr. Alban, is trained on my own work and continuously evolves to stay current with the AI landscape — built on the same stack the companies ship to clients.
// PERSONAL AGENT
Ask it about my work, my approach to regulated AI, or how a specific product is architected — it answers the way I would, and gets sharper over time as my work evolves.
[ 10 ] // EDUCATION
// DEGREE
University of Oslo
// DEGREE
BI Norwegian Business School
// DEGREE
// CERTIFICATIONS
[ 11 ] // EXPERTISE
// AI
// Architecture
// Security
// Cloud & DevOps
[ 12 ] // CONTACT
For collaboration, advisory or technical deep-dives — pick the channel that fits.
[email protected]
// PHONE
+47 900 39 911
// ADDRESS
Langesgate 8
3210 Sandefjord
Norway
in/hakon-berntsen →