TeacherHub

Language-learning platform with a CEFR-aligned LLM-as-judge eval suite over its homework-generation pipeline.

I started learning Spanish and wanted my teacher to have better tooling than a shared Google Doc. TeacherHub is the result: a language-learning platform built solo, in production for its two users, my Spanish teacher and me, since early 2026.

Teachers schedule and run video lessons with a live notepad for capturing errors and vocabulary as they happen. After the lesson, an AI pass turns those notes into a summary, an updated skill graph, and homework suggestions. Homework itself is generated by a three-stage pipeline (plan, generate, verify) and graded against the CEFR level of the student. Students get a dashboard, offline-capable homework, and a spaced-repetition vocabulary game.

TeacherHub teacher dashboard showing the lesson schedule, an action-required banner, and tabs for students, library, and lesson types The teacher dashboard: schedule, review queue, students, and content library.

The part I'd show a hiring manager first is the eval suite: 68 LLM-as-judge scenarios across three AI pipelines, each scored on a five-dimension rubric, with a human-review log that recalibrates the judge prompt when human and judge scores diverge. See the Evals & testing tab.

By the numbers: 22 documented features, 101 database migrations, 20 accepted ADRs, and a monorepo that ships to Vercel and to iOS/Android through Capacitor.

Links

  • GitHubPrivate repo. Ask me and I'll grant view access.