Victor Herbst · educator, linguist & developer

I build learning systems that know where the learner gets stuck.

I connect pedagogy, language, AI and product engineering to turn complex learning problems into experiences people can actually use.

Céu Canto practice completion screen
Céu Canto contextual feedback screen
real product · mobile first
14+years inside online communities
1,000+learning objects in production
54/54external production checks
01

flagship product

Céu Canto · English learning platform

A teaching philosophy turned into a production system.

Course, guided reading, curated decks, spaced review, teacher monitoring and live-class experiences — designed as one coherent learning journey rather than a pile of features.

Product strategyLearning designFull-stack engineeringAI workflows
Céu Canto reconstructed wordmark
Even the name is a product decision: language, poetry and machine legibility living in the same mark.
Visit the live product

Selected systems inside Céu Canto

One product, four kinds of hard problem.

01

Contextual practice

Feedback explains not only what was wrong, but why the learner's interpretation happened — then schedules the right return instead of an arbitrary repetition.

pedagogy · interaction · spaced review
02

A living language library

Curated decks, guided reading and a structured grammar path share one vocabulary layer, so content compounds instead of becoming isolated lessons.

content systems · reading · curriculum
03

Teacher visibility

The product treats the teacher as part of the system: learner journeys, classroom practice and intervention points are designed together.

teacher tools · learning data · operations
04

Production discipline

Payments, offline review, privacy, migrations, release gates and external smoke checks turn an educational idea into a dependable service.

engineering · quality · launch

Original product engines

Not a skin. A set of reusable systems.

ENGINE 01 · CHARACTER SYSTEM

Twenty pets with a runtime of their own.

Each character is a semantic 48 × 48 grid with multiple authored states, weighted idle paths and a safe renderer. Motion pauses in hidden tabs and respects reduced-motion preferences.

  • 20 original characters
  • Discrete, cancellable animation loops
  • Palette patches by semantic region
Four real Céu Canto pixel pets rendered from the product engine
ENGINE 02 · OKLCH COLOR MATH

Seeds become a complete, accessible interface.

One to five colors enter Prisma. A deterministic solver produces 13 semantic roles, derived states and a trace — with gamut mapping, collision rules and WCAG AA contrast judged on the emitted colors.

3 seeds
3 validation passes
Atmosphere#240B05
Surface#32140B
Primary text#F8E7E2
Noun phrase#945736
Adjunct#31447A
Target#895F22
Accent#9CB679
Identity#3E5289
Positive#90DAB9
Negative#DE6573
ENGINE 03 · ADAPTIVE ATMOSPHERES

Twenty-four skies, paired across day and night.

Every sky is a governed token system for background, surfaces, text, actions, identity and learning heatmaps — not a loose color preset. The product stores allowlisted IDs, never freeform CSS.

Noitenight
Alvoradanight
Sertãonight
Matanight
Vaga-lumenight
Marénight
Algodãoday
Tangerinaday
Girassolday
Lagoaday
Amanhecerday
Lavandaday

12 shown · 24 production skies · 12 paired families

Applied intelligence

AI is useful when it improves a decision.

A better next activity. A clearer teacher intervention. A content workflow the team can trust. A process that stops wasting time. The model is never the product; the improved decision is.

Learner

Personalized paths and feedback grounded in observable behavior.

Teacher

Tools that clarify where attention and human judgment matter most.

Team

AI-assisted content and operations with standards, review and traceability.

Business

Reliable funnels, instrumentation and workflows that improve with use.

How I work

From teaching insight to working software.

  1. 01

    Find the real learning problem

    Start with the learner's mistake, the teacher's constraint and the behavior we want to change.

  2. 02

    Model the pedagogy

    Translate linguistic judgment into content rules, data structures, interactions and measurable signals.

  3. 03

    Ship a complete slice

    Design, implement and connect the smallest experience that can be used by a real person end to end.

  4. 04

    Learn from reality

    Use feedback, product evidence and teaching outcomes to make the next iteration more precise.

Portrait of Victor Herbst

Teacher mind. Builder hands.

I am most useful where language, people and systems collide.

I have taught language since 2020, built digital products from the ground up and lived inside online communities since 2012. That mix lets me move from a grammar explanation to a product flow, from a content idea to an operational system, without losing the human voice along the way.