Denkkappe
An independent consultancy for AI and data science you can understand, trust, and audit.
Get in touchWe build high-performance data science models where every outcome can be understood, trusted, and audited. For organizations that need to meet regulatory standards, win stakeholder buy-in, or accelerate research without betting on a black box.
Our work spans regulated finance, biotech research, and language processing at scale, from glass-box credit scoring to signal and image analysis. Science you can trust, and put to work.
Most “smart” business ideas die a quiet death. Not because the market rejected them, but because nobody was brave enough to tell the founder their baby was ugly.
If you’re spending weeks tweaking pitch decks in isolation, stop navel-gazing and bring the idea to us. Expect your assumptions dismantled, your machine learning approach stress-tested, and your ego slightly bruised.
It isn’t a casual gut check. It’s a rigorous diagnostic built on decades of academic research and enterprise strategy, tailor-made for every idea. We’re not here to validate your feelings; we’re here to validate your viability.
Every pitch deck now says “AI”. Before you back one, find out whether the AI is real, defensible, and built on data the company actually has the right to use.
Matteo works hands-on: he reviews the code and the data room, interviews the technical team, and tests every claim against what he finds. In particular:
You get a written technical assessment, the red flags ranked by severity, and the questions to put to the founders before you sign. Conducted personally and under NDA, typically before a term sheet.
Strictly AI and machine learning: a technical assessment, not investment advice.
It looks at the AI at the heart of your idea: the science, the data, and the models it depends on. Every idea has to survive five questions.
Before any model: does the underlying physics, chemistry, or biology support the idea, and is the signal you’re betting on detectable in data you can realistically get?
Or would a simpler, older, more transparent method do the job better? Sometimes the honest answer is no AI at all.
Where it comes from, who owns it, whether there’s enough of it, and what biases it carries into your model.
Especially when users are vulnerable. A general-purpose chatbot that charms in a demo can say things no responsible product should, and regulators are paying attention.
If your product is a thin layer over someone else’s model, what stops a competitor, or the model’s maker, from doing the same next month?
A frank working session on your idea, your deck, and any data or prototype you have. Then a written verdict: go, fix this first, or stop, with the reasons, the evidence, and what would change the answer.
Your idea stays yours. Past sniff tests have ranged from subsurface resource exploration to assistive technology, and every detail stays with its founders. NDA on request.
Book a sniff testAI won’t save a bad concept. It will just help you build a bad product faster.
Glass-box models whose reasoning can be inspected, explained, and audited, not just benchmarked.
The full toolbox of statistical learning, simulation, and network science. Large language models when they’re the right tool, not by default.
A first-principles understanding of complex systems, paired with the hands-on skill to ship production-grade machine learning pipelines.
Honest verdicts and measurable outcomes over vanity metrics. As Voltaire put it, perfect is the enemy of good.
A few of the problems Matteo has tackled over the years, anonymized.
A digital lender needed credit-scoring models that were accurate, and also explainable to regulators and to the businesses applying. Matteo built glass-box scoring models, fraud and anti-money-laundering checks, and the production pipeline to run them, together with a data team he built from scratch. The models scored more than 100,000 applications, behind billions of euros in financing.
A fintech startup set out to automate treasury management for companies. Matteo designed its AI engine from the ground up and helped shape it into a product investors would back. The venture raised seed funding and was generating revenue within two years.
Real estate investors wanted a systematic way to spot unconventional deals before the market did. Matteo led the development of algorithms that found them in messy, heterogeneous data, and delivered pilots to leading real estate firms.
A global fact-checking effort needed AI that makes human fact-checkers faster without taking judgment out of their hands. Matteo directed the large-scale language-processing research behind it, and turned that research into production systems that supported people rather than replacing them.
A regional emergency medical service had to cut costs without slowing its response. Matteo built a simulation of the whole system, ground and helicopter ambulances included, showing where spending could be rationalized and how to deploy vehicles under a tighter budget.
Denkkappe is Matteo Morini’s consultancy, based in Lugano.
Matteo has moved between research and industry for more than twenty years. He spent over a decade as a tenured research scientist at the University of Turin, did his doctorate in computer science at the École Normale Supérieure de Lyon along the way, and co-authored books on agent-based economics published by Cambridge University Press and Palgrave Macmillan.
In industry, he built the data teams and explainable credit-scoring models behind billions of euros in financing at an Italian fintech, and today works as a partner and advisor to ventures in fintech, biotech, and large-scale simulation. More than a decade of teaching postgraduates means he can explain complex things plainly.
Watching founders slap an “AI” label on matrix multiplication over a frozen latent space, as if fifty years of statistical learning and pre-LLM research never existed, is officially my villain origin story.
Matteo
Whether it’s a model that has to hold up to an auditor or an idea that has to hold up to reality, tell us what you’re working on.
Email Matteo