NeuroWeave: Your Personal AI University

NeuroWeave: AI that takes you from zero to mastery in 7 days. Not a chatbot—a Marathon Agent that plans your curriculum, teaches with voice+visuals, tests you, and syncs to Drive. Learn anything.

Comment

NeuroWeave: Your Personal AI University

Inspiration

We've all experienced it: you want to learn quantum physics or blockchain, so you watch three YouTube videos, skim Wikipedia, bookmark a course. A week later? Nothing sticks.

The problem isn't access—it's structure. Online courses have 90%+ dropout rates. Chatbots answer questions but vanish immediately after. Traditional education demands rigid schedules.

We asked: What if an AI could architect your entire learning journey, teach through voice and visuals, draw alongside you on a whiteboard, and remember everything across days?

That's NeuroWeave—a Marathon Agent that doesn't sprint through single responses but runs the full distance with you, from curiosity to mastery.

What It Does

NeuroWeave transforms vague goals like "learn artificial intelligence" into structured 7-day learning odysseys.

Smart Curriculum Generation

Enter any topic—or photograph a textbook page. NeuroWeave builds a visual learning map: a node-based graph where each concept unlocks only after prerequisites are mastered. No more learning things out of order.

Multimodal Lessons

Every lesson combines:

  • Adaptive written explanations that adjust to your level
  • AI-generated diagrams custom-made for each concept (Imagen 3)
  • Motion graphics for dynamic ideas—watch atoms bond or neural networks learn (Veo 3.1)
  • Real YouTube videos and sources via Google Search grounding—zero hallucinated links

Live AI Tutor with Interactive Whiteboard

This is the breakthrough.

Activate Tutor Mode. Ask, "Can you explain derivatives?" The AI speaks the explanation while drawing the curve live on a shared whiteboard. You draw your attempt. The AI sees your sketch and corrects it in real-time: "Almost! But the tangent line should touch here..."

It's like having a brilliant tutor sitting beside you, red pen ready, infinitely patient. Low-latency voice makes interruptions natural. Multimodal vision lets the AI see and respond to your drawings.

For hands-free learning, Voice Command Mode lets you navigate lessons and solve problems entirely through speech.

Adaptive Testing

Learning without feedback is just reading. NeuroWeave verifies understanding through:

  • Dynamic Quizzes: Questions adapt in real-time. Ace it? Harder questions. Struggling? Remedial micro-lessons appear.
  • "Reveal Me" Challenges: The AI generates an abstract image. You identify the concept. Did you truly understand "photosynthesis," or just memorize words?
  • Voice Logic Puzzles: Hands-free Socratic riddles. "If you're at light speed and turn on headlights, what happens?"
  • Whiteboard Problem-Solving: Solve math/physics problems while the AI watches your approach and guides you.

If you struggle, NeuroWeave backtracks automatically—offering simpler explanations, analogies, or prerequisite refreshers.

Persistent Memory

Unlike chatbots, NeuroWeave remembers everything:

  • Smart Notes: AI-generated summaries in Markdown
  • Session Transcripts: Complete tutor conversation records
  • Progress Analytics: Mastery levels, time per topic, quiz performance
  • Visual Library: All diagrams and whiteboard drawings catalogued

Everything auto-saves to Google Drive. Close your laptop Tuesday, open your phone Thursday—pick up exactly where you left off.

How We Built It

Multi-Agent Orchestra

Instead of one model doing everything poorly, we orchestrated five specialized Gemini models:

Agent Model Purpose
Curriculum Architect Gemini 3 Pro (Thinking Mode) Deep reasoning (32K tokens) to structure pedagogical dependencies
Live Tutor Gemini 2.5 Flash (Native Audio) Real-time voice + vision + whiteboard drawing via tool use
Visual Synthesizer Imagen 3 Custom educational diagrams—not stock images, tailored illustrations
Motion Designer Veo 3.1 Short animations for dynamic concepts (reactions, motion, flow)
Quiz Master Flash Lite Lightning-fast adaptive questions maintaining learning flow
Resource Finder Google Search Grounding Real YouTube links, papers, articles—zero hallucinations

Technical Highlights

Deep Thinking for Curricula: Gemini 3 Pro uses 32,768-token reasoning budgets to genuinely contemplate pedagogical structure—not just pattern-match.

WebSocket Real-Time Magic: Persistent connections enable <200ms voice latency. When the AI decides to draw, it calls create_visual_aid—we intercept, trigger Imagen client-side, render on canvas. All while voice continues seamlessly.

Multimodal Vision Streaming: Your whiteboard sketches stream as image frames alongside audio. The AI sees what you draw continuously—not screenshots, but live visual awareness.

Progressive Loading: Text renders instantly. Images lazy-load (2-3s). Videos are opt-in (8-12s). Users feel constant progress, never frozen screens.

Rolling Summarization: Long sessions compress older messages into semantic summaries every 10 exchanges. Maintains continuity, respects context limits.

Stateful Drive Sync: OAuth 2.0 + Drive API v3. Markdown notes upload silently after each lesson to a dedicated folder. Your learning history lives across devices.

Tech Stack

React 19, Vite, TailwindCSS, Google GenAI SDK, Google Drive API, Web Speech API, Lucide Icons.

Challenges We Ran Into

The Latency Paradox

Deep thinking (3-5s) + image generation (2-3s) + videos (8-12s) = 20+ second waits. Users quit.

Solution: Progressive rendering. Show text instantly, images lazy-load, videos are opt-in. Skeleton loaders with animations keep the UI alive. Users see constant progress.

WebSocket vs React Lifecycle

Components unmount, connections drop, audio buffers vanish, history disappears.

Solution: Context Provider with ref-based state. Audio queues in persistent buffers. Connection state lives outside components. Conversations never break.

Drawing Synchronization

When AI draws, Imagen takes 2-3 seconds. Early versions froze awkwardly.

Solution: Optimistic UI. Show "AI is drawing..." animation immediately. Voice continues explaining. Image fades in when ready. No dead air.

Context Exhaustion

After 15-20 minutes, conversation history exceeds limits.

Solution: Semantic compression, not truncation. Distill older exchanges into summaries preserving insights. Users never notice; AI remembers what matters.

API Quota Burn

Veo limits drained in hours during testing.

Solution: Client-side caching (reuse videos), visible "Video Budget" UI, opt-in generation. Scarcity makes videos feel valuable.

Accomplishments That We're Proud Of

True Multimodal Understanding

We didn't just bolt voice onto text. The AI sees your drawings, hears your questions, and responds with speech + visuals simultaneously. It's the closest thing to a real tutor we've ever built.

Deep Reasoning That Works

Using Gemini 3 Pro's thinking mode with 32K-token budgets produces curricula that feel designed by experienced educators, not regurgitated topic lists. The quality leap is night and day.

Purposeful AI-Generated Media

Every diagram, animation, and video serves learning—not spectacle. We learned to integrate media exactly when concepts need visualization. Users report genuine "aha!" moments.

Zero Hallucinations

Google Search grounding transformed trust. Every YouTube link works. Every citation checks out. Users stopped questioning references.

Persistent Memory

Most AI forgets you the moment you close the tab. NeuroWeave remembers your entire journey—progress, struggles, breakthroughs—across days and devices.

Gamified Learning That Doesn't Feel Childish

The node-unlock system, adaptive difficulty curves, "Reveal Me" challenges—they create genuine engagement without patronizing adult learners.

What We Learned

About AI

  • Thinking Mode is transformative for structured tasks. Extended reasoning produces pedagogically sound curricula, not just topic dumps.
  • Multimodal streaming is brutally hard but worth it. The user experience of talking to an AI that sees and draws feels magical.
  • Grounding eliminates hallucination anxiety. Users trust the system completely when every link works.
  • Specialization beats generalization. Five focused models outperform one jack-of-all-trades.

About Users

  • Progressive revelation beats information dumps. Unlocking one day at a time transformed completion rates.
  • Voice needs visual feedback. Waveforms and live transcription make speech feel reliable.
  • Adaptive difficulty creates flow states. Too easy = boredom. Too hard = frustration. Dynamic adjustment keeps learners engaged.
  • AI-generated media must justify itself. Every image needs a "why now" answer, or it's just decoration.

About Education

  • Structure matters more than content. The same information, sequenced intelligently, produces 10x better outcomes.
  • Accountability drives completion. Testing, progress tracking, and visible advancement combat dropout.
  • Multimodal teaching reaches more learners. Visual + auditory + kinesthetic options let students find their optimal path.

What's Next for NeuroWeave

Collaborative Learning Rooms

Enable groups to share curricula, compete on leaderboards, and teach each other. The AI moderates peer explanations, ensuring accuracy.

AR Tutoring

Imagine the AI drawing diagrams in your physical space via AR glasses or phone AR. Study anatomy with a 3D heart floating on your desk.

Verified Certification

Partner with universities and institutions to offer blockchain-verified credentials for completed learning sprints. Real proof of mastery.

Emotional Intelligence Layer

Use sentiment analysis to detect frustration, boredom, or confusion. The AI adapts teaching style in real-time—switching from technical to conversational, or suggesting breaks.

Offline Mode with Local Models

Download curriculum packages and use on-device AI for learning anywhere—flights, remote areas, data-limited regions.

Mobile-First Experience

Native iOS/Android apps with optimized voice interfaces for learning on the go. Commute time becomes study time.

Specialized Domains

Pre-built expert tracks: "Medical School Foundations in 30 Days," "Software Engineering Bootcamp," "Financial Markets Mastery." Curated by domain experts, taught by AI.

Integration with Real Courses

Sync with Coursera, edX, or Khan Academy. NeuroWeave becomes your personal tutor for existing courses, filling gaps and answering questions 24/7.

NeuroWeave isn't just an app—it's proof that AI can democratize world-class education.

In a world where AI can reason deeply, speak naturally, generate custom media, and remember context over days, the traditional classroom model becomes optional.

Everyone deserves a brilliant tutor. NeuroWeave is our first step toward that future.

Learn anything. For real. In 7 days.

Built with: React 19, Google GenAI SDK (Gemini 3 Pro, Gemini Live API, Imagen 3, Veo 3.1, Flash Lite), Google Drive API, Web Speech API.

Built With

Share this project:

Updates