An app-agnostic language relationship

Your course already lives
where you talk.

Message it. Send it a voice note. Open a generated exercise only when you need one. The method follows your life instead of asking you to enter another app.

3+chat channels
1learning brain
5adaptive formats

The app is the relationship—not the icon

Learn where your conversations already live.

Text or send a voice note. The same learning engine understands the turn, answers naturally and opens a tiny personalized workspace only when a screen helps.

Telegram is implemented first. WhatsApp and Instagram use the same normalized event contract.

P
probandolenguastelegram · learning coordinator
Cuéntame quién eres. Puedes escribir o mandar un audio. Los errores no importan todavía.
0:12voice note

Browser simulation; the Telegram adapter accepts real voice-note files.

Phone context · 8 minutes

Social flow

Learning CoordinatorLocal fallback · approved atoms only

Initial plan compiled locally from approved atoms.

Current focushola · me llamo · ¿y tú? · soy de · me gusta

01 / INTRODUCTIONSCC0
Hola, me llamo Kavi.Hi, my name is Kavi.

Your turn · 10 seconds

Say your name, your place, one thing you like.

Imperfect is welcome. Nothing is recorded or uploaded.

01A real introductionlistening + recognition
02Send a ten-second hellospeaking + conversation

Evidence, not engagement

Listening is not writing.
Exposure is not mastery.

Every mode touches the same approved atoms. What changes is the evidence it can honestly produce.

Kavi / introductionsToday
Listening13%
Writing0%
Speaking3%

Confirmed handwriting updates this tab only. Passive plays: 0; exposure never raises mastery.

CC0

Rights are a runtime constraint.

Corpus atoms stay rights-verified. Ephemeral tutor replies are generated from the learner’s turn and are never silently added to the corpus.

  • 5,500 project-authored CC0
  • 1,000 modern Chile / LatAm
  • 300 Chile-specific constructions

The small-universe function · E(L)

Communicate more with
the language you know.

Start with 50 human-curated atoms, then add ten at a time. Mathematics audits the foundation and selects expansions against balanced CC0 conversational nuclei; it does not replace the method’s foundation.

Semantic evaluation: preregistered · not run
First +10 pack

algo · antes · gustar · hoy · más · otro · plan · poder · preferir · querer

19%
50
25%
60
28%
70
30%
80
32%
90
33%
100

Balanced conversational reach · 48 held-out nuclei at N=100 · semantic evaluation remains unrun