Every idea can be taught many ways.
Most learning platforms pick one explanation and hope it lands. Embris finds the many explanations that already exist, measures how each one teaches, and gives you the one that fits how you learn.
A non-generative learning platform, built in the Netherlands.
How it works
Three ideas, and the whole platform follows from them.
Many explanations, one idea
The same concept taught visually, rigorously, by worked example. No learner is stuck with the one explanation that didn't click.
Matched to how you learn
Embris measures the character of each piece of content and each learner's preferences across interpretable dimensions, then ranks by fit. The platform re-personalises around you as it learns more.
An AI that guides, not generates
Axi locates and cites existing material and refuses to write explanations. No invented content, everything attributed at its source.
Axi
We built an AI for learning that refuses to explain anything.
Ask Axi to explain eigenvalues and it won't. Not because it can't — because we made it decline. When an AI writes an explanation, four things quietly stop working: you can't check it against a source, you can't attribute it to whoever did the thinking, you can't tell whether it's the kind of explanation that works for you, and you can't tell whether it's right.
So Axi only points at real material and cites it. If it can't find something real, it says so. Refusal is a feature we built on purpose — and it's enforced in code, not policy.
Pilots
Two universities, two countries, September 2026.
Embris is being trialled with real courses and real students, starting with Challenge-Based Learning at TU Eindhoven.
Content is surfaced link-out with attribution at the source. Embris harvests only from curated open repositories via their APIs, publishes a machine-readable AI/TDM opt-out, and does not train models on learner data.
Join us
We're five people covering six areas of work.
There's a working prototype, two pilots landing in September, and more open ground than we can cover. If you'd rather own an area outright than pick up tickets, we should talk.
Frontend
A node-graph learning-line designer with real interaction problems and nobody owning the surface.
Infrastructure
The entire production foundation, greenfield, on EU infrastructure, with a September deadline.
AI & machine learning
Measuring how content teaches across interpretable dimensions — and building the evaluation set that can tell whether it works.
Business
Incorporation, funding routes, and turning two pilots into a company.
Technology law
DSA, AI Act, licence compatibility, and a genuinely novel contributor-licence question.
Communications
No press kit, no voice, no website beyond this page. Build it from zero.
Being straight about the terms. These are unpaid founding-contributor roles. We're pre-revenue and pre-incorporation, forming a Dutch BV, and the founder split isn't settled — so we won't quote you an equity number we can't back.
What we can say is that the people who join before incorporation are the people that decision gets made among. If that's the trade you want, write to us and tell us which area interests you.