VIDEO: https://www.loom.com/share/40732cd7131c42c1a87b0a9a7bc2e7a9

AI-Powered Personalized Learning System

A complete pipeline that transforms voice input into fully personalized courses, powered by AI at every step.

What This System Does

1. Listens to what you want to learn (voice onboarding)
2. Understands who you are as a learner (AI profiling)
3. Creates a learner brain (unified profile)
4. Generates a personalized course structure (AI-driven)
5. Delivers customized lessons (coming next)

The Key Insight

Same subject + different learner = different course

Traditional courses are one-size-fits-all. This system generates a unique course for EACH learner based on:

• What they want to learn
• WHY they want to learn it
• Where they're starting from
• What they already know
• What they need to skip
• How deep they want to go
• Their learning style and tone preference

Quick Start

Setup

# Install dependencies
cd brain && npm install
cd ../course && npm install

# Set API key
echo "OPENROUTER_API_KEY=your_key" > .env

Generate a Complete Brain + Course

# Option 1: Use existing data from onboarding/profiling agents
cd brain
npm run generate  # Creates brain from agent outputs

cd ../course
npm run generate  # Generates course structure

# Option 2: See it work programmatically
cd course
npm test

Verify Personalization

cd course
npm run verify

This generates courses for two different learners learning the same subject. If personalization works, they should be meaningfully different.

System Architecture

Voice Input
    ↓
┌─────────────────────────────────────┐
│  Phase 1: Voice Onboarding          │
│  "I want to learn X because Y"      │
└────────────┬────────────────────────┘
             ↓
     onboarding_result.json
     { subject, why }
             ↓
┌─────────────────────────────────────┐
│  Phase 2: AI Profiling              │
│  Deep questions to understand       │
│  learner identity                   │
└────────────┬────────────────────────┘
             ↓
     learner_profile.json
     { level, gaps, strengths, depth... }
             ↓
┌─────────────────────────────────────┐
│  Phase 3: Brain Creation            │
│  Combine onboarding + profiling     │
└────────────┬────────────────────────┘
             ↓
     user_brain.json
     { high_level: { onboarding, profiling } }
             ↓
┌─────────────────────────────────────┐
│  Phase 4: Course Generation (AI)    │
│  Generate phases + lessons based    │
│  on COMPLETE learner profile        │
└────────────┬────────────────────────┘
             ↓
     course_structure (in brain)
     { phases: [...], lessons: [...] }
             ↓
┌─────────────────────────────────────┐
│  Phase 5: Teaching (NEXT)           │
│  Deliver personalized lessons       │
└─────────────────────────────────────┘

Project Structure

.
├── onboarding-agent/          # Phase 1: Voice onboarding
│   ├── index.js
│   └── onboarding_result.json
│
├── profiling-agent/           # Phase 2: Identity profiling
│   ├── index.js
│   └── learner_profile.json
│
├── brain/                     # Phase 3: Brain management
│   ├── createBrain.js         # Combines onboarding + profiling
│   ├── updateCourseStructure.js
│   ├── generate-brain.js      # CLI tool
│   └── user_brain.json        # The complete brain
│
├── course/                    # Phase 4: Course generation
│   ├── generateStructure.js   # AI course generation
│   ├── structurePrompt.js     # System prompt
│   ├── generate.js            # CLI tool
│   ├── test.js                # Tests
│   └── verify-personalization.js  # Proves personalization works
│
└── SYSTEM_OVERVIEW.md         # Detailed architecture docs

Example Output

Input: User Brain

{
  "high_level": {
    "onboarding": {
      "subject": "history of modern tech founders",
      "why": "to gain practical business knowledge for my startup"
    },
    "profiling": {
      "starting_level": "complete beginner",
      "depth": "solid working knowledge",
      "focus_areas": "fundraising, product-market fit, tactical lessons",
      "skip_areas": "biographical details, pre-modern history",
      "learner_context": "Current startup founder"
    }
  }
}

Output: Personalized Course

{
  "title": "Founder's Playbook: Strategic History of Modern Tech Titans",
  "phases": [
    {
      "phase_number": 1,
      "title": "The Dawn of the Digital Ecosystem",
      "lessons": [
        {
          "lesson_number": 1,
          "title": "Identifying Opportunity: Netscape and the Consumer Internet",
          "description": "Study Marc Andreessen's market creation",
          "instructional_seed": "Focus on finding early signals rather than technical specs"
        }
      ]
    }
  ]
}

Notice how the course is:

• Titled for a founder (not generic)
• Focused on strategy (not biography)
• Starts with opportunity identification (actionable)
• Uses founder-relevant examples (Netscape, not academia)

How Personalization Works

Every field in the brain matters:

Field	Effect on Course
subject	What the course is about
why	Shapes the angle and focus
current_level	Where to start (skip basics or start from zero)
skip	Topics that won't appear
teach_from_scratch	Topics that get dedicated lessons
depth	How granular, how many lessons
gaps	Weaknesses that need addressing
strengths	Things to build on, not re-teach
background	Shapes examples and framing
tone	Affects titles and style

Testing Personalization

The key test: Do different learners get different courses?

cd course
npm run verify

This generates two courses for the same subject but different learners:

• Learner A: Beginner PM, wants practical tools only
• Learner B: Technical PM, wants deep AI product knowledge

The structures should be meaningfully different in:

• ✅ Course titles
• ✅ Phase topics
• ✅ Lesson granularity
• ✅ Tone and terminology
• ✅ What's included vs. skipped

API Requirements

OpenRouter API Key required for:

• Identity profiling (Gemini)
• Course structure generation (Gemini)

Add to .env:

OPENROUTER_API_KEY=your_key_here

Get a key at: https://openrouter.ai

What's Built

✅ Phase 1: Voice onboarding ✅ Phase 2: AI identity profiling ✅ Phase 3: Brain creation and management ✅ Phase 4: AI-powered course structure generation

What's Next

🚧 Phase 5: Teaching agent

• Read course structure
• Generate actual lesson content from instructional_seed
• Deliver lessons interactively
• Track mastery and write to brain.low_level

🚧 Phase 6: Adaptation engine

• Read performance data from brain.low_level
• Adjust future lessons based on struggles
• Personalize delivery dynamically

Key Design Principles

1. Brain is the single source of truth - Everything reads from it
2. Dynamic profiling - AI decides what to ask based on the learner
3. Complete context - Course generation sees onboarding + profiling together
4. Instructional seeds - Structure guides future content generation
5. Testable personalization - Different profiles = different courses

Documentation

• SYSTEM_OVERVIEW.md - Complete architecture and data flow
• brain/README.md - Brain structure and usage
• brain/INTEGRATION.md - How phases connect
• course/README.md - Course generation details

Running the Full Pipeline

# 1. Onboarding (run manually or use existing data)
cd onboarding-agent
node index.js

# 2. Profiling (run manually or use existing data)
cd profiling-agent
node index.js

# 3. Create brain
cd brain
npm run generate

# 4. Generate course
cd course
npm run generate

# 5. View result
cat ../brain/user_brain.json

Result: A complete user_brain.json with:

• high_level filled (onboarding + profiling)
• course_structure filled (personalized phases + lessons)
• low_level empty (ready for teaching phase)

Example Use Cases

Use Case 1: Product Manager Learning AI

• Subject: "AI tools for PMs"
• Profile: Non-technical, beginner, wants practical only
• Result: 5-phase course on tool selection, workflows, no theory

Use Case 2: Founder Learning Tech History

• Subject: "History of tech founders"
• Profile: Startup founder, wants tactical lessons
• Result: 4-phase course on strategy, fundraising, PMF (no biography)

Use Case 3: Developer Learning Product

• Subject: "Product management"
• Profile: Engineer, wants to transition to PM
• Result: Course that leverages technical background, focuses on business skills

Each learner gets a completely different course structure, even if the subject is similar.

Contributing

See individual module READMEs for development details:

• brain/README.md - Brain system
• course/README.md - Course generation

License

MIT