WellVersed

WellVersed — a complete English teaching platform for online teachers and schools: interactive lessons, homework, and voice-rich slides, ready to teach at every level.

Comment

Accomplishments that we're proud o## What it is

WellVersed is a high-tech edtech platform that gives English teachers a complete, classroom-ready teaching system — interactive lessons, matching homework, and voice-rich slides — for every CEFR level from A1 to C2. Teachers don't build materials or wrangle tools. They open the app and teach. Everything already feels handcrafted, sequenced, and premium.

Inspiration

Good English teachers spend more time making lessons than teaching them — hunting for texts, building slides, writing homework, recording audio, leveling everything for the right learner. The craft gets buried under production work.

I wanted to flip that: a platform where the entire curriculum is built for the teacher, so the only thing left to do is the human part — teaching. The teachers using it don't think about the machinery underneath. They just experience lessons that feel like a specialist designed them by hand.

What it does

  • Generates a full curriculum tree — Course → 12 topics → 8 lessons each — with hidden pedagogical seed material carried between each stage so nothing drifts off-topic.
  • Assembles each lesson from 53 interactive slide templates spanning warm-up, listening, reading, vocabulary, grammar, pronunciation, speaking (including free voice recording), writing, and review.
  • Produces a matching homework deck for every lesson, staged from supported practice toward independent communication.
  • Adds AI-generated imagery and voice audio so listening and pronunciation slides come alive without a studio.
  • Runs everything through a multi-stage automated review pipeline (pedagogy, language, interactivity, and structure) before a lesson is ever imported.

How I built it

The platform is a React + TypeScript + Vite front end on a Firebase backend (Firestore, Cloud Functions, Hosting, Storage), with a slide-based lesson engine and a strict JSON lesson schema that keeps every generated slide compatible with the active templates.

The generation pipeline is model-orchestrated: a planning pass structures the lesson, generation passes fill the templates, and multiple review passes audit pedagogy and interactivity before import. It's built to run across several model providers, and I'm bringing OpenAI models in to strengthen the reasoning-heavy stages — curriculum planning, lesson sequencing, and the quality-review passes where language precision and pedagogical judgment matter most.

Media is handled by dedicated image and text-to-speech generation so each slide ships with real visuals and audio, not placeholders.

What I learned

  • Structure beats prompting. The wins came from the curriculum architecture — seed details flowing between stages, a fixed template registry, staged review — far more than from any single clever prompt.
  • Invisible AI is the goal. The best outcome is a teacher who never once thinks "this was AI-made." That bar forced real quality gates instead of raw generation.
  • Consistency is a systems problem. Making every screen and every level feel like "the same hand designed it" is a design-system and validation challenge, not a styling one.

Challenges I faced

  • Keeping generation on-rails. Free-form output constantly wants to invent its own lesson. Hidden seed material + a strict template schema + multi-stage review were the fix.
  • Quality at scale. A lesson that's 90% right is still not classroom-ready. The automated pedagogy/language/interactivity review pipeline exists to close that last gap.
  • Media that matches the lesson. Generating images and audio that actually fit the slide's teaching intent — not generic stock — took dedicated prompt discipline.

What's next

OpenAI credits go directly into the pipeline's brain: deeper reasoning for curriculum planning and sequencing, sharper language review, and richer generated media — so WellVersed becomes the best-in-class way to build and teach English, end to end.

Built With

Share this project:

Updates