LuminaOS

Every student deserves a personal tutor. LuminaOS uses Gemini 3's native audio to deliver voice-first Socratic tutoring at scale — with full teacher transparency.

Comment

We kept asking: why doesn't every student have a personal tutor?

Research shows 1-on-1 tutoring produces a 2 standard deviation improvement in learning outcomes — known as Bloom's 2 Sigma Problem. But hiring a tutor for every student is economically impossible.

Existing EdTech tools bolt AI onto the side — a chatbot in a sidebar that students ignore and teachers can't monitor. When Google released Gemini 3 with native audio, we realized we could build something fundamentally different:

An AI that doesn't just type at students — it speaks to them, draws for them, and reports to their teachers. All in real-time.

LuminaOS is an AI-native education operating system where Gemini 3 is a first-class participant in learning.

For Students

  • 🎙️ Lumina Sync — Voice-native tutoring. Speak naturally with a Socratic AI tutor that listens, responds in voice, and draws diagrams on a live whiteboard. After each session, Gemini 3 compiles resources via Google Search Grounding.
  • 💬 Lumina Chat — Text-based AI tutoring with real-time streaming and automatic topic synthesis.

For Teachers

  • 📊 AI Analytics Hub — Every student-AI interaction appears instantly. Topics, engagement scores, transcripts, and AI grading insights — all in real-time via Firestore onSnapshot.
  • 👨‍🏫 Teacher Assistant — Gemini 3-powered tools for drafting announcements, generating quizzes, and analyzing class performance.

Built with React 19 + TypeScript, powered by four distinct Gemini 3 capabilities:

Capability What It Does
Gemini 3 Native Audio Real-time voice tutoring via WebSocket — no TTS, the AI thinks in audio
Gemini 3 Flash Chat AI, grading analysis, topic synthesis, teacher tools
Gemini 3 Image Gen Generates educational diagrams on a live canvas whiteboard
Google Search Grounding Discovers real resources (Khan Academy, YouTube, Wikipedia) after sessions

Firebase serves as the entire backend:

  • Firestore — Real-time persistence with onSnapshot listeners
  • Firebase Auth — Google Sign-In + email/password
  • Firebase Hosting — CDN-backed deployment

11 Gemini 3 integration points across 6 component files.

  • Native audio streaming — Manually encoding PCM to base64, managing WebSocket lifecycle, handling interruptions, and synchronizing AudioBufferSourceNode playback timing without clipping or overlap.
  • Real-time whiteboard generation — Detecting visual concepts during voice sessions, generating images without interrupting audio flow, and rendering on canvas asynchronously.
  • Firestore composite indexes — Required for real-time analytics queries (ordering sessions by timestamp within classrooms). Had to be manually created and took time to build.

  • React 19 + TypeScript strictness — New type definitions required esModuleInterop configuration and explicit type annotations throughout.

  • ✅ Voice-first AI tutoring works — natural spoken conversation with Gemini 3 that feels like a real tutor

  • ✅ The AI draws while it talks — Gemini 3 generates a parabola on the whiteboard while explaining quadratic equations

  • ✅ Zero-latency teacher awareness — sessions appear on the teacher dashboard the instant they finish

  • ✅ 11 Gemini 3 integration points — woven into every layer: voice, text, vision, search, grading, content creation

  • ✅ Production-deployed — live at academy-94e78.web.app with demo credentials for instant access

  • Gemini 3's native audio is a game-changer — the quality difference vs. text-to-speech is immediately noticeable

  • Real-time systems require thinking about every state transition — connections opening, closing, interrupting, reconnecting

  • Firebase onSnapshot makes real-time features almost trivially easy once the data model is right

  • The Socratic method is surprisingly well-suited to AI tutoring — prompting Gemini 3 to ask questions rather than give answers increases student engagement dramatically

  • 🌍 Multi-language tutoring — Gemini 3's native audio supports multiple languages

  • 📈 Long-term progress tracking — aggregate sessions into learning trajectory curves

  • 📝 Assignment-aware tutoring — Lumina tailors guidance to the specific assignment a student is working on

  • 👥 Collaborative voice sessions — multiple students in one Lumina Sync session for group study

  • 🔒 Cloud Functions proxy — move API calls server-side for production security

  • 📱 Mobile PWA — responsive design optimized for phones and tablets

Built With

Share this project:

Updates