AI for regulated industries
I help banks, lenders, insurers and asset managers build, advise on and lead AI that reaches production and holds up when risk, compliance and the regulator come asking.
Built and shipped inside
01 The problem
The demo is the easy 10%. The invisible 90% is where AI projects stall: integration, data governance, model risk, explainability, audit trails, sign-offs, change management. In a regulated business that 90% is bigger, slower, and the cost of getting it wrong is existential.
Impressive proofs-of-concept that quietly die in the gap between a notebook and production.
Promising automation blocked at the approval gate because risk was never designed in from day one.
Models nobody can explain to an auditor, a regulator, or your own second line of defence.
02 Is this you?
Most of the people I work with arrive in one of these situations.
An impressive proof-of-concept that cannot get past integration, model risk or sign-off.
KYC, onboarding, credit, claims or document work done by hand, at a cost you can count in full-time staff.
Every initiative dies at the approval gate because compliance was an afterthought, not a design input.
03 Why me
Most people are one or the other: AI builders who have never survived a model-risk review, or compliance advisors who could not ship an agent to production. I have spent twelve years on both sides. The value lives in the overlap.
AI that ships to production and passes audit.
04 How we can work together
From building the systems, to advising the boardroom and the cap table, to leading the team that runs it all.
I embed and build the capability.
Fractional AI leadership that stands up the team, operating model and culture, so the capability outlasts me.
I help you decide.
Strategy and diligence for the leaders and investors making high-stakes AI calls.
I build the AI.
Turning manual, compliance-bound processes into AI systems that reach production and survive audit.
How the Audit works
Week 1
Interviews and a walk-through of your highest-cost manual processes and the constraints around them.
Week 2
I map the automation opportunities, the ROI, and the compliance and risk that gates each one.
Week 3
A board-ready readout: prioritised, sequenced, costed and de-risked.
05 Proof
90%+
AML/KYC & credit decisioning automated
Defined and delivered AML/KYC and credit-decision automation across retail and SME segments in Finland, Sweden and Denmark, replacing manual onboarding end to end. Pioneered the compliance-first delivery model the bank adopted organisation-wide. An earlier credit-automation design took SME automation from 5% to 50% and added ~€1M in annual profit.
~1M CHF
Annual value from enterprise AI
Led enterprise AI transformation recovering 20,000+ editorial hours a year. Shipped an AI content platform handling 380K+ events annually and a RAG assistant that lifted session duration 39% for 2M+ monthly users.
Acquired
Drove product-market fit to a successful exit
Took an inherited MVP for explainable AI in fixed-income asset management to product-market fit, directly enabling acquisition by BondIT. Explainability treated as a first-class, regulator-ready feature, not an afterthought.
$100K → $4.5M+
ARR as Head of Product
Owned the product vision and roadmap that scaled ARR 45x. Built the GPU-compute and MLOps offering, forged partnerships with Intel Habana and NVIDIA, and landed the largest enterprise contract at the time (€300K+).
06 Testimonials
“Stefan is one of the brightest guys I've worked with. He went often beyond his role as product manager to also get his hands dirty in operations and architecture, and took on communicating complex and sometimes difficult messages to executives, and orchestrating complex operations, often dealing with compliance, with stakeholders across multiple organizations.”
“Stefan is an expert in Artificial Intelligence and High-Performance Computing, and a remarkable professional. His expertise spanned both software and hardware, going way beyond the level of depth I would typically expect from a Product Manager. He was always reliable, consistently meeting our milestones, and I genuinely enjoyed working with Stefan and his team.”
07 Investment advisory
Funds and founders bring me in for technical diligence on AI deals: reading the team, the architecture and the defensibility before the money moves. The same judgement that ships production AI is what tells the signal from the hype.
4.02x
Net MOIC on Groq (117% IRR, post-carry)
Selected AI investments advised
Groq
AI inference · SAFE
Replit
AI coding · Series D
Together AI
AI cloud · SAFE
Galaxia AI
Credentials
08 About
AI Product Leader & Strategy Consultant
I started in bank risk and credit, building ML scoring pipelines and automating lending decisions inside regulated FinTechs and at Danske Bank, where I learned that in finance the model is the easy part. Getting it past risk, compliance and audit is the real work.
From there I led AI product at Scorable (explainable AI, acquired), scaled Germany's first AI cloud platform at Genesis Cloud, and ran enterprise AI transformation at Ringier worth around a million Swiss francs a year. I still build hands-on: my own research-grade platform, ProtocolEngine, runs eleven specialised AI agents in production.
That mix is the whole point. I am equally comfortable fine-tuning a model and sitting in a model-risk committee, and I help regulated businesses capture the upside of AI without the downside.
Speaking & community
09 Teams & culture
The fastest way to waste an AI investment is to make it depend on one person. I build high-performing teams and the culture to keep them shipping, so when the engagement ends the capability stays.
10 Insights
Talk · AIPMX, 2026
The demo is the easy 10%. Governance, evaluation, model risk and change management are the invisible 90%, and they are where most AI projects quietly die.
Talk · Generative AI World Summit & MLOps World
How a well-scoped fine-tune on open models can beat foundation models on a specific task, at a fraction of the cost and with full control of the weights.
Talk · MLOps Community
Where large-model training breaks down across many GPUs, and the optimisation work, with MIT/ISTA researchers, that fixes the bottlenecks.
11 FAQ
I work inside your environment and controls: your cloud, your data boundaries, your model-risk process. Nothing leaves your perimeter without sign-off, and governance is designed in from day one rather than bolted on.
Yes, that is the point. I have shipped AML/KYC and credit automation through a bank's second line of defence. I design for explainability, audit trails and sign-off so your risk and compliance teams can approve rather than block.
Two to three weeks. You get a prioritised set of automation opportunities ranked by ROI and risk, the compliance constraints mapped per use case, and a business case you can take to the board.
Both. I am hands-on: I fine-tune models, build RAG and agentic systems, and set up MLOps. I also lead strategy and embed as a fractional AI lead. Use me for any point on that spectrum.
Regulated, compliance-heavy businesses: banks, lenders, insurers, asset managers, and increasingly health and pharma. The common thread is high stakes and a low tolerance for getting AI wrong.
Get started
Tell me about the manual process that is costing you the most. If an Audit is a fit, I will come back within two business days with how it would work. If it is not, I will tell you that too.
Book a 20-min intro call→Prefer to write? Use the form, or the details below.