Vidya Verse

Type any concept, step inside it — AI-generated VR science lessons in your language.

Comment

Vidya Verse

Type any concept, step inside it — AI-generated VR science lessons in your language.

Vidya Verse turns any science topic into a fully explorable, narrated 3D VR micro-lesson in under a minute. A student or teacher types a prompt like "Class 10 — Human Heart cross section", picks a language, and the app drops them inside a life-size WebXR scene with labeled hotspots, a friendly AI tutor, narrated voiceover, and an interactive drag-and-drop quiz.

It runs on a ₹8,000 Android phone with Google Cardboard, on a laptop as a 3D orbit viewer, and on a Meta Quest 3 for room-scale VR. No app install. No content team. No headset gatekeeping.

Inspiration

We grew up in classrooms where the human heart was a 2D line-drawing on a dusty blackboard. Meanwhile, private schools across the world were handing out VR headsets with pre-built science scenes. We built Vidya Verse because we wanted every teacher in Bharat — especially in Tier-2/3 and rural schools — to have a generative science lab in their pocket, not a locked library of Western scenes behind a ₹30,000 headset.

What it does

A teacher or student types any science concept in Hindi or English, and within about a minute Vidya Verse drops them inside a narrated, explorable 3D VR lesson built specifically for that topic. You can click parts of the model for labeled callouts, ask the AI tutor follow-up questions with your voice, take a drag-and-drop quiz, and share the lesson as a short URL that works on a phone, laptop, or Meta Quest — no app install, no account.

How we built it

Next.js 14 App Router with Tailwind on the front, A-Frame 1.6 for the WebXR scene graph, and a single route handler that chains Google Gemini 2.0 Flash (lesson scripting + voice Q&A) to Meshy / Tripo (text-to-3D) and ElevenLabs / Web Speech (TTS). Lessons serialize to a JSON scene graph we cache in localStorage, which makes them shareable, remixable, and fully offline after the first load.

Challenges

Text-to-3D APIs regularly take 60+ seconds and occasionally return ugly meshes — so we built a hand-tuned A-Frame primitive library (heart, atom, solar system, cell, brain, flower) that acts as a beautiful fallback and keeps the demo visual quality high even with zero generation credits. Second, making the voice-Q&A loop feel instant: we cap LLM responses to 1–2 sentences and stream the TTS so perceived latency stays under 3 seconds.

Accomplishments

A single URL that works identically on a ₹8,000 Android phone, a MacBook, and a Quest 3. Hindi-first narration with on-the-fly language toggle. Every MVP feature on the brief — prompt-to-lesson, AI tutor Q&A, hotspots, quiz, cross-device WebXR, bilingual narration, teacher dashboard, API-key panel — is implemented and working in this repo.

What we learned

The bottleneck for generative XR in classrooms is not the model — it is the orchestration and the fallback path. A great demo means pre-caching, graceful degradation, and a scene-graph schema cleanly decoupled from any one 3D vendor. Also: A-Frame is still the fastest way to ship a WebXR app that 'just works' across devices in 2026.

What's next

1) Lip-synced VRM avatar with real visemes (Gabber.dev pattern) 2) Tamil, Kannada, Bengali narration 3) Teacher roster with per-student progress synced via Supabase 4) AR Quick Look export for iOS classrooms 5) Offline-first PWA with IndexedDB scene caching for low-bandwidth schools

Built for the DesignXR Hackathon by Aryan Choudhary.

Built With

Share this project:

Updates