Learning Systems Architect · EdTech DACH · German C1

Your AI gives feedback
on grammar students
haven't studied yet.

That is not a prompt problem. It is an architecture problem.

In 2024, students in Week 2 of German A1 were receiving corrections on Perfekt and Akkusativ — structures they had not seen. The fix was not a better prompt. It was a typed pedagogical domain (WochenKontextSchema via Zod) that enforces what the AI can and cannot correct at each instructional stage.

The AI does not decide. It obeys the domain.

See the architecture huhugerman.com — live evidence Start a conversation

Domain-Driven Design TypeScript · Supabase · Zod DaF / German L2 AI Governance DACH EdTech C1 German · Offenburg 2019 15 Years · UAM · Real Students Scopus Peer Reviewer

The intersection

Three domains. One problem that requires all three.

🎓

DaF Pedagogy

15 years teaching German A1–C1 at UAM Azcapotzalco, ENP UNAM, FFyL UNAM. Metacognitive scaffolding. Authentic materials from A1. Hamburg fieldwork (2019). CONACYT fellow. Two institutional curricula designed and adopted.

🔬

Research Methodology

M.Ed. Pedagogy UNAM (2020). B.A. German Literature UNAM (2015). 11 formal peer-review contributions in Scopus-indexed journals (2017–2025). Vandergrift, Gadamer, Krashen, Borg — applied in a live classroom, not cited in isolation.

⚙️

Systems Engineering

B.Sc. Web Systems Development, UdeG (GPA 98.5, 2025). Domain-Driven Design. Zod schema contracts. ADRs. Supabase · PostgreSQL · TypeScript · Astro · Google Apps Script v8.

The problem with AI in EdTech: a pedagogue without engineering builds methods that don't scale. An engineer without domain knowledge builds systems that hallucinate pedagogy. A researcher without a classroom produces theory without calibration.

This profile solves problems that break when any one of the three is missing.

Featured Project

huhuGERMAN

A pedagogical platform and research system. Not a toy project — used by 30–40 real students per trimester at UAM Azcapotzalco since 2022.

Production ● Live

Domain-Driven German Language Learning Platform

Architectural Decision Record (ADR)

The pedagogical domain is formalized as a typed Zod schema. The AI doesn't decide; it obeys.

// Pedagogical Domain Contract
export const WochenKontextSchema = z.object({
  gesehen:       z.array(z.string()),  // structures taught
  nicht_gesehen: z.array(z.string()),  // not yet taught
  korrektur:     KorrekturSchema      // what AI can / cannot correct
});

huhuGERMAN operationalizes the distinction between global comprehension and selective comprehension. The AI is architecturally forbidden from correcting grammar the student has not yet studied. Pedagogy governs the system. Not the other way around.

Astro TypeScript Zod Supabase PostgreSQL Google Apps Script v8 DeepSeek AI Domain-Driven Design

→ huhugerman.com → CSE Diagnostic Demo

Architecture Decision Record

Why the ADR arrived late — and what that teaches

The huhuGERMAN ADR was written after the MVP, not before. That honesty is deliberate. A system that documents its decisions retroactively understands something a system that never documents them never will.

2022 — MVP

Endpoint + AI + implicit context

Validated the flow: student submits → AI gives feedback. Proved the pedagogical hypothesis. But context lived in the teacher's head, not the system. The AI corrected Perfekt for students in week 2. System not governable.

2024 — Friction Discovery

Domain-level misalignment surfaces

Students reported confusion from feedback on grammar they hadn't seen. This was not a bug. It was a missing domain contract. "When a system depends on human judgment to not break, it's not a product."

2025 — ADR Written

Pedagogical domain as typed source of truth

WochenKontextSchema formalizes gelernt / nicht_gelernt / korrektur. prompt-builder constructs prompts from domain. AI obeys, not decides. Zod validates at every boundary. JSONB snapshots enable future audit.

2026 — Academic Publication

Targeting Die Unterrichtspraxis / System (Elsevier Q1)

The system becomes evidence for the research trajectory: thesis (2020) as diagnostic phase, huhuGERMAN as intervention phase.

Pedagogical Decision Record

PDR: Governing the AI, not just building it

If an AI doesn't know the "State of the Classroom," it shouldn't generate a single line of content. The PDR is a hard-coded contract—ensuring pedagogy governs technology, not the other way around.

The Contract

Pedagogical Domain as Typed Truth

huhuGERMAN formalizes the "WochenKontext"(Weekly Context). No generic LLM magic. If a concept (like Plurals for A1) is marked as "out_of_scope" in the PDR, the system is architecturally forbidden from generating it.

The Governance

PDR as System Prompt

Every prompt is wrapped in a PDR snapshot. The AI obeys the boundary constraints defined by the educator. The error is not a bug; it's a boundary condition.

RFC-001

Standardizing Pedagogy

Moving from "I hope the AI gets it right" to "The system cannot generate it wrong." The PDR structure is being open-sourced to push the EdTech industry

15 Years of Verifiable Evidence

From field research to production-grade systems

Three degrees. 11 formal peer-review contributions in Scopus-indexed journals (2017–2025).

UNAM · 2020 · CONACYT

M.A. Pedagogy

Research: Comparative German pedagogy (Mexico–Germany). Peer-reviewed publication. This served as the diagnostic research phase that established the pedagogical requirements for huhuGERMAN.

UNAM · 2015

B.A. German Language & Lit.

Thesis: Metacognitive strategies for listening. First formal application of Vandergrift’s framework at CELE-UNAM. Foundation for the platform's comprehension architecture.

UdeG · 2025

B.Sc. Web Systems Dev.

Systems Architecture: GPA 98.5. The technical bridge. Converts complex pedagogical domain logic into executable Zod schemas and production-grade contracts.

Universität Offenburg · 2019

German C1 Certification

Domain Expertise: Specialist in the Hochdeutsch–Umgangssprache gap. Validated fieldwork to ensure AI feedback aligns with real-world language usage, not just textbook theory.

Does your EdTech platform have this problem?

If your AI gives feedback on content students haven't studied yet — or if you are building German language learning for the DACH market and need domain expertise that spans pedagogy, research, and engineering — that is the specific intersection I work in.

Available for instructional design, curriculum architecture, AI training data (German C1), and solutions engineering for EdTech DACH. Remote

Email yassergandhi.dev@gmail.com LinkedIn /in/yassergandhi GitHub /yassergandhi Live evidence huhugerman.com

Your AI gives feedbackon grammar studentshaven't studied yet.