CEMI Labs · Learning
Learning, tested live.
This is where CEMI experiments with how people learn. Every project on this page is a live experiment you can experience today — opened on a date, tested with real learners, and written up honestly, including what it taught us and what it changed.
The lab, narrated — every experiment on this page, in order.
Education is about the learning experience, not the teaching experience. Everything this lab builds is judged from the learner’s side of the desk: what a person can now understand, do and become — never how efficiently something was delivered.
That is why our methodology is called Smoother Experiences and not courses, why the aiLearning Challenge weighs process over product, and why our coaches gauge a learner before they teach anything. A tool earns its place here only when it deepens the time a learner spends with the learning object.
If we try to make change look like the past or respond to the expectations and structures of the past, it’s not true change, is it?
Carlos Miranda Levy, Coordinator of Impact, Innovation and Intelligence
Each entry is a running experiment. The date is when it opened — on this site, or earlier where the work predates it; the proof is the thing itself.
What we are testing — The five aspects of the Experience Layer — Wonder, Hook, Play, Disruption and Horizon — declared at every altitude and never scored; Disruption run as a declared experiment with a written stop rule, one per Learning Experience, bounded and followed by consolidation; and Horizon held as a value the class holds, not a lever it pulls — a lever only when tied to a concrete next step the learner can take without asking permission. All of it a design position authored by Carlos Miranda Levy, never validated pedagogy.
What it has taught — How it is being verified, and what that cost: a deep-research round run in September 2026 across eighteen areas of literature, every source checked one by one against the primary text or the publisher’s own record, four ledgers and a Gate 2 synthesis — 129 deduplicated rows. What the round found: several of the sources that one of three commercial deep-research systems supplied did not exist — a resolving-looking DOI on a paper that was never written, a page locator for a quotation that is not on the page — which is why nothing entered the canon until a human had read it at its source. Completeness of fields is a risk signal, not a quality signal: the return that arrives with every field filled in is the one most in need of checking.
See the program (in Spanish)What we are testing — A learning-first methodology built around the learning object, the participant and an orientation role — applied first to onboarding, where a new hire’s first weeks become a designed learning experience.
What it has taught — Vocabulary is a design decision: participant, never student; learning object, never syllabus. The rule protects the methodology’s own voice and now governs every CEMI learning product.
Read the methodology, document by document
What we are testing — The methodology implemented as a live online campus — hundreds of ready-to-run programs on AI and beyond, open to the public, with custom campuses for organizations.
What it has taught — A methodology only counts once strangers can enroll in it. The public campus is the lab’s standing proof that Smoother runs at scale, not only on paper.
Enter the campus
What we are testing — Describe what you want someone to learn and the Designer generates a learning unit or a content item with the Smoother methodology — built on the three pillars: the learning object (what is learned), the learning subject (who learns) and the real environment (where it will be applied) — ready to review, download and share.
What it has taught — A learning experience begins with three questions, not with content. When the object, the subject and the environment are named first, the AI’s output changes from generic material into a design someone can actually run.
Open the Designer
What we are testing — A live competition of creating with AI, from primary school to university: sealed briefs a generic AI answer cannot finish, a divergence gate, process weighed over product, and a final defense with no AI in the room. Beside it, the Challenge Gym: an unscored practice layer that never qualifies anyone, defended by construction — a decoy demonstration, a decoy task and a containment clause.
What it has taught — A full rehearsal with twenty educators on 18 August 2026 produced six lessons — and each one changed a rule you can read, with its reason, in the evolution log.
Read the rehearsal lessons
What we are testing — Play as a learning technology: four games from our own ecosystem — ArcLycée — Aventura Arqueológica Dominicana, La Travesía de Cary, Solar System Simulator, CEMI Fútbol — free, in the browser, each built so the fun is inseparable from the learning object.
What it has taught — Engagement is not the goal; it is the mechanism. The games that hold attention longest are the ones where the rules of the game are the rules of the subject.
Play the games
What we are testing — Adaptive tutors for real exams and subjects — Brevait, our French-writing coach for the brevet, first among them. Each coach gauges the learner’s level before teaching anything, and never needs an account to start.
What it has taught — Anonymous-first is a learning decision, not only a privacy one: the first conversation happens before any barrier, and the coach earns the sign-in by being useful.
Meet the coaches
What we are testing — A standing council of learning perspectives — personas inspired by history’s great educators, contemporary practitioners, and the learners and parents education answers to — each carrying a full pedagogical stance.
What it has taught — Authority is earned by what a perspective thinks and values, never by a claimed title. The panel’s personas are openly fictional and argue from stance, lineage and conviction.
Meet the panel
What we are testing — 29 long-form articles, in three languages, with narrated films and audio editions — the lab’s written record, from the maps we copied as children to the expired syllabus of 2026.
What it has taught — Ideas ship with an expiry date. The 2026 catalogue of useful AI flaws in The Expired Syllabus is explicitly due for revision in 2027 — the essay practices what it preaches.
Read the Insights
What we are testing — The lab’s first experiment: a guided builder that turns an educator’s intention into a working prompt, with AI refinement for signed-in users.
What it has taught — The earliest lesson still holds — a good prompt is a design of the learning task, not a request to a machine.
Build a promptThree findings that came out of the experiments above — each with the record it lives in.
When any artifact can be generated in seconds, only the learner’s time with the learning object proves anything. Every activity now belongs to one of four families — what AI cannot do, what requires it, where it fails, and what is processed after it with the tool off.
The Expired SyllabusThe rehearsal confirmed the corpus: a room full of different ideas drifts toward one average idea unless a gate forces three substantially different approaches before anyone builds. The Divergence Gate is the counter-measure.
Challenge foundationEvery rule change in the Challenge is logged with its date and its why, and every rehearsal lesson is shown with what was decided and implemented. A format that keeps its promises in public is easier to trust — and to improve.
Evolution logThe newest Insights — the place the lab thinks out loud.
The lab works with people who bring real learners and real questions.
Schools, universities and training organizations redesign their learning experiences with us — from a single activity to a whole campus.
Work with usHost a rehearsal with your educators using the open kit, or bring the aiLearning Challenge to your school or district.
Open the rehearsal kitA learning problem the lab should test, a game, a coach, a panel voice — tell us what you would put on the bench.
Write to the labFour labs, one conviction: technology earns its place by amplifying people.
The Learning Innovation Lab is operated by aiLearning.global for CEMI Labs, under the Smoother methodology.
