{{ role.title }}
You own{{ role.ownsCap }}.
{{ role.intro }}
The role in full
What you will do
- {{ pt.lead }} {{ pt.rest }}
What you bring
- {{ x.t }}
Nice to have
- {{ x.t }}
Success after six months
- {{ x.t }}
{{ nodeBody }}
How we work
First principles decide what we build. A flexible foundation keeps it useful as learning, and AI, keep changing.
We start from what learning is, a mind moving between states of understanding at a cost in effort, not from what learning apps already look like.
Learning is a search from what you know to what you want to understand.
A feature earns its place only if it widens the routes a learner can take, or lowers the cost of finding a good one.
Where we are unsure, real learners’ journeys settle it, not our guesses.
Nobody knows what learning will need in five years, so to stay future-proof we stay flexible. We build like a language: a few words and rules that can express unlimited ideas.
Where we apply it
Every learning step is an action, defined abstractly, so new kinds can be added.
Pages are made of blocks, and a block can hold anything. New block types are easy to add.
Tools act on a page, from a calendar to a one-click AI skill. New ones plug in the same way.
The usual catch, and how templates solve it
Flexibility usually makes new things harder to build. Templates fix that: ready-made combinations of blocks, tools and actions you can use as they are or change, the way Ubuntu and Mint package Linux.
Why this makes Embris future-proof
A new step, block, tool or AI skill is one more piece. Nothing already built breaks.
Templates give a complete setup on day one; every part stays adjustable.
Anyone can make blocks, tools and templates, so Embris is not limited to what we build.
A flexible foundational system is a general-purpose base designed to support many structures and uses, including ones its creators have not anticipated.
Foundational: other things are built on top of it. System: its parts follow shared rules and work together. Flexible: parts can be added, removed, rearranged or extended without rebuilding the base.
The rule we design by: define a small set of stable building blocks and rules, then let complexity emerge from how they are combined.
Design commitments
Axi points to rated, human-made explanations. Generative AI is a clearly marked last resort.
Linked and credited where it was published, never copied.
Embris adds routes around a course instead of replacing it.
The platform will run on EU infrastructure under the GDPR. Data is only ever shared anonymised and aggregated.
Communities
The World Community will be open and free to everyone. Universities, courses, companies and research groups can run private communities, and release their work to everyone when it is ready.
Institutions bring the material they already have. No rewriting or migration project needed.
A real-world experiment
Which routes reach understanding fastest, and which learning styles really exist.
Gaps show what to write next. Ratings show what works.
Anonymised, aggregated evidence that teaches AI to explain better. Never personal data.
Investors, universities, partners and future teammates are all welcome.
Open roles
We are three students building Embris in Eindhoven. These are founding roles with real equity and real ownership, not jobs.
Terms
These roles are unpaid for now. This is what you get and what we ask.
{{ t.lead }} {{ t.rest }}
The roles
You own{{ role.ownsCap }}.
{{ role.intro }}
We do not assume these two are the complete list. If you see an area you would be good at that we have not named, tell us.
Pitch it by emailWhat the area is, why it matters to Embris specifically, what you would own, and what you would have done by the end of the first pilots. One page is plenty.
Design and the learner-facing experience, content partnerships with educators and publishers, and community: the people who rate and annotate explanations.
Tell us which role, what you have made before, and what you think we are getting wrong. We reply to everyone.
Or message any of us on LinkedIn:
Work with us
A mission-scale problem, one product, five revenue lines.
We are raising a pre-seed round. Ask us for the deck.
Talk to the foundersA private space for your courses or internal knowledge, built from material you already have.
Discuss a subscriptionTry Embris with one cohort and see where learners get stuck. Pilots are planned from November 2026.
Propose a pilotYour work stays where you published it, credited to you. Publishers will share in a royalty pool based on usage.
Become a partnerTwo founding roles are open, and we welcome pitches for others.
See the open rolesFive students with the same goal can need five different explanations. Embris finds each of them the one that fits, from everything already published.
The argument from first principles
Every search begins from what the learner already knows.
The goal state is whatever they are trying to understand.
Between the two lie actions.Reading a chapter, watching a lecture, attempting a problem or asking a colleague each moves the learner to a new state of understanding.
A sequence of such actions constitutes a solution.A student working on eigenvectors might watch a lecture, attempt three problems, ask a peer where the reasoning failed, then read a single worked example.
Two things can be improved: the actions available, and the cost of finding a good route.
1. Widening the search space
In practice a learner chooses among the few options immediately in front of them.
The reachable field is considerably larger:every state those actions could produce, and every state reachable from those in turn. Most of it remains invisible.
Embris therefore connects the existing field rather than reproducing it.Indexing the actions that already exist yields a wider search space than any single publisher can assemble.
A single action does not leave every learner in the same place.Initial states differ, and so do the transitions an action produces. Each reader is left needing something slightly different.
Annotations supply the missing resolution.Because they are small, they can close a gap from anywhere in that spread, and because they are shared, each gap need only be closed once, by whoever encountered it first.
2. Reducing the cost of search
Learning is paid for twice:once to establish which sequence works, and once to walk it.
The first cost need only be paid once, after which it can be inherited by everyone who follows.
Embris lowers that cost with learning lines, ranking, learning styles and Axi.
Over time, what learners actually do on Embris shows which ways of teaching work.
See our solution How we work
Why this matters now
AI already writes code, generates data and evaluates other AI. Once that loop runs at machine speed, capability can outpace our understanding: formally in control, practically unable to steer.
The intelligence that finds a new antibiotic can find a worse pathogen. Which one we pursue is a human choice, and choosing well takes understanding.
Many informed people decide better than a few experts (the Condorcet Jury Theorem). Below the threshold of understanding, the effect flips: more voices multiply the error.
Not so everyone understands everything, but so humanity stays close enough to the frontier to choose where intelligence looks.
Roadmap
{{ p.what }}
Investors, universities, educators, or anyone an explanation has failed: write to us.
{{ mailAddress }}Write to Embris
To {{ mailAddress }}
Subject: {{ mailSubject }}