00:00 / 30:00
1 / 24
System Overview

🎓 Universal Adaptive Learning System

100% LLM-Powered Education with 100 Intelligent Tutoring Agents

100 ITS Agents
45 Methods
3,375 Learning Combinations
17 Languages

Speaker Notes

Welcome the audience. Introduce UALS as a next-generation adaptive learning platform built entirely on LLM technology. Emphasize that this is production-ready software, not a research prototype.

Key points: 100 intelligent agents, 45 total methods (15 KE + 15 SPL + 15 SATA), 3,375 unique learning combinations, 17 supported languages.

Duration: 1-2 minutes

📋 Presentation Agenda

Part 1: Foundation ~10 min

  • The Doctor-Teacher Analogy
  • Multi-Philosophy Framework
  • 15×15×15 Personalization

Part 2: Learning Systems ~10 min

  • Knowledge Explorer (KE)
  • Socratic Playground (SPL)
  • Scenario-Based CAT (SBCAT)

Part 3: Intelligence ~8 min

  • 100 ITS Agent Architecture
  • 28 Software Agents
  • Explainable AI Pedagogy

Part 4: Platform ~7 min

  • Teacher Workflows
  • xAPI Analytics
  • Technical Architecture

Speaker Notes

Walk through the agenda. Mention that timing is flexible - 30 min for overview, up to 60 min with deep dives.

🏥 The Doctor-Teacher Analogy

A Framework for AI-Enhanced Education

Medical Practice
UALS Education
Patients receive diagnosis
Students receive learner model assessment
Prescription based on symptoms
Content recommendation based on knowledge gaps
Side effects monitored
Cognitive overload & frustration detected
Treatment adjusted over time
Real-time pedagogical adaptation
Evidence-based medicine
Evidence-based pedagogy (+0.30 to +0.90 SD)

Speaker Notes

This analogy helps stakeholders understand why AI in education requires the same rigor as AI in medicine.

🎯 Dual Philosophy, Unified Platform

Supporting Multiple Educational Approaches

📘

Curriculum-Based

Domain → Subdomain → Concept

  • ✓ Structured learning sequences
  • ✓ Predefined progression
  • ✓ Group-paced instruction
  • ✓ Standardized assessment
Best for: K-12, certifications, compliance training

Same UI • Same Features • Same AI Support

🎖️

Competency-Based

Category → Competency → Proficiency Level

  • ✓ Student-driven paths
  • ✓ Self-paced exploration
  • ✓ Mastery-based progression
  • ✓ Personalized goals
Best for: Professional development, skills training
📚 Full Documentation

Speaker Notes

Critical architectural achievement: two fundamentally different philosophies on one platform without code duplication.

🧮 The 15×15×15 Framework

3,375 Unique Learning Combinations

15 Knowledge Explorer Methods

Diverse ways to explore content: concept maps, problem-based scenarios, case studies, simulations...

×
15 Teaching Strategies

Evidence-based pedagogies: Socratic dialogue, EMT feedback, cognitive apprenticeship...

×
15 Assessment Formats

Next-gen SATA formats: classic, priority ranking, weighted confidence...

= 3,375 Unique Personalized Learning Paths

Speaker Notes

This is NOT 3,375 pre-authored content paths – it's dynamically generated combinations.

🎓 Three Integrated Learning Systems

🔍

Knowledge Explorer (KE)

Deep-dive exploration with dynamic concept mapping

15 methods Visual learning Schema building
💬

Socratic Playground (SPL)

Interactive tutoring with 5 agent types

15 pedagogies EMT framework Adaptive dialogue

Scenario-Based CAT (SBCAT)

Adaptive testing with IRT analytics

15 SATA formats Continuous flow Per-competency tracking

All three systems work in BOTH curriculum and competency philosophies

Speaker Notes

These three systems form the core learning experience. Click any card to open detailed documentation.

🔍 Knowledge Explorer (KE)

15 Methods for Content Exploration

📊 Linear Tabs
🔄 Progressive Disclosure
🗺️ Concept Map
📚 Tabbed Accordion
🎯 Focus & Expand
⏳ Timeline View
🔗 Linked Cards
📖 Book Format
🎬 Presentation Mode
❓ Q&A Guided
🧩 Jigsaw Discovery
🔬 Case Study
📈 Compare/Contrast
🎮 Interactive Simulation
🌳 Tree Explorer
🧠 AI adapts density based on learner profile
🔗 Dynamic concept relationships
📱 Responsive across devices
📚 Explore All 15 Methods

Speaker Notes

Each method is grounded in cognitive load theory and multimedia learning principles.

💬 Socratic Playground (SPL)

15 Evidence-Based Pedagogical Approaches

+0.82 SD

Socratic Dialogue

Inquiry-based questioning

+0.75 SD

EMT Framework

Expectation-Misconception-Tailored

+0.74 SD

Reciprocal Teaching

4 Cs: Clarify, Question, Summarize, Predict

+0.65 SD

Cognitive Apprenticeship

6 methods: Modeling to Exploration

+0.62 SD

Self-Explanation

Deep learning through elaboration

+10

View all pedagogies

5 SPL Agent Types:

🎓 Tutor 💡 Hint Socratic 📚 Study Mate 💬 Feedback
📚 Full Pedagogy Documentation

Speaker Notes

Effect sizes are from meta-analyses in educational research.

Scenario-Based CAT (SBCAT)

15 Next-Generation SATA Assessment Formats

📋

Classic SATA

Multiple correct answers

📊

Priority Ranking

Order by importance

⚖️

Weighted Confidence

Certainty scoring

🔗

Conditional Logic

If-then reasoning

⏱️

Time-Pressured

Timed responses

🧩

Matrix SATA

Multi-dimensional

Continuous Flow

No rounds - unlimited questions cycling through competencies

Per-Competency Tracking

Individual metrics for each skill area

Rolling Window Scoring

Based on most recent N questions

📚 All 15 SATA Formats

Speaker Notes

SATA formats reduce measurement error by 30-40% compared to traditional MC.

🤖 100 Intelligent Tutoring Agents

The Complete Adaptive Learning Ecosystem

30 Production-Ready Core
Learner Model Knowledge Tracing Pedagogical Model Adaptive Sequencing Feedback Generation +25 more
27 Extended Capabilities
Transfer Learning Cognitive Load Cultural Adaptation AR/VR Immersive Accessibility +22 more
43 Micro-Behavior Monitoring
Mouse Patterns Keyboard Dynamics Pause Analysis Scroll Behavior Attention Detection +38 more

🎯 Learning Insight Synthesis Agent (The 100th Agent) - Synthesizes data from all 99 monitoring agents

📚 Explore All 100 Agents

Speaker Notes

This is the most comprehensive ITS agent architecture in educational technology.

🧠 Agent Category Deep Dive

📚

Knowledge Representation

Domain Model, Curriculum Alignment, Misconception Detection

3 agents
👤

Learner Modeling

Knowledge Tracing (BKT/PFA/DKT), Mastery Estimation, Affective State

4 agents
🎯

Pedagogical Decision

Socratic, EMT, Reciprocal Teaching, Cognitive Apprenticeship

9 agents
💬

Tutoring Interaction

Dialogue, Feedback, Hints, Scaffolding

5 agents
📝

Content Generation

Problems, Worked Examples, Explanations

4 agents
📊

Assessment

Performance Analytics, xAPI Tracking

2 agents

Speaker Notes

Walk through the 6 main categories. Emphasize Knowledge Tracing uses BKT, PFA, or DKT.

⚙️ 28 Software Infrastructure Agents

Production Software That Powers UALS

🔐 Security & Auth

  • Cookieless Auth Agent
  • OAuth Integration
  • Session Management
  • Role-Based Access

💾 Data & Caching

  • GCS Content Cache
  • MD5 Hash Deduplication
  • LRS Cache Fallback
  • Version Management

🤖 LLM Gateway

  • Multi-Provider Support
  • Prompt Orchestration
  • Token Optimization
  • Fallback Handling

📊 Analytics

  • xAPI Statement Builder
  • LRS Integration
  • Real-time Dashboards
  • Report Generation
94-96% Token cost reduction through intelligent caching
📚 Software Architecture Docs

Speaker Notes

These agents work behind the scenes to ensure the 100 pedagogical agents can focus on learning.

🧠 Explainable AI Pedagogy

"Show AI Thinking" - Real-time Decision Transparency

1 Learner Submits Answer
2 Agents Analyze
3 Decisions Made
4 Content Generated

10 Analysis Depth Levels:

Level 1: Lightning (4 agents, 1-2s)
Level 2: Quick (11 agents, 2-4s)
Level 5: Professional (50 agents, 12-18s)
Level 10: Complete (100 agents, 25-40s)
⏯️ Playback controls 📊 Visual timeline 🎨 Color-coded events 🔒 GDPR compliant (5-min cache)

Speaker Notes

The Workflow Visualizer shows exactly which agents were invoked and what decisions were made.

🛡️ AI Content Safeguards

5 Layers of Human-Verified Learning

5

Institutional Oversight

Admin approval workflows, audit trails

4

Content Expert Review

Subject matter expert verification

3

Teacher Editing

Version control, edit tracking, approval

2

GCS Content Cache

Reviewed content served, not raw LLM output

1

LLM Quality Filters

Prompt engineering, output validation

📚 Safeguard Documentation

Speaker Notes

Students NEVER receive raw, unreviewed LLM output.

👩‍🏫 Teacher Workflow

From Framework to Classroom

1️⃣

Browse Frameworks

Select from school DSC catalog or create custom

2️⃣

Copy to Personal

Create personal copy for customization

3️⃣

Add AI Notes

Customize content generation instructions

4️⃣

Create Class

Assign framework, set context

5️⃣

Invite Students

Share class link or code

🔄 Role-based permissions 👥 Content expert collaboration 📝 Ownership transfer 📊 Full audit trails
📚 Class Management Guide

Speaker Notes

Key architectural principle: LRS = Pointer (dscId), GCS = Data (actual competencies).

📊 xAPI Learning Analytics

Every Interaction Tracked for Insights

15 Standardized Verbs:

answered completed explored interacted mastered attempted achieved viewed +7 more

Analytics API Endpoints:

  • /api/xapi-analytics/competency-performance
  • /api/xapi-analytics/sata-analysis
  • /api/xapi-analytics/report-card
  • /api/xapi-analytics/learning-history
  • /api/xapi-analytics/class-analytics
🔒 Customer-owned data in their LRS
📈 Real-time dashboards
🎯 Per-competency proficiency tracking

Speaker Notes

CRITICAL: Never fetch all statements (could be millions). Always use targeted queries.

🏗️ Technical Architecture

Presentation Layer

Universal Templates • Dynamic DOM • Responsive UI

↓ ↑

Application Layer

Express.js • Philosophy Detection • Route Handlers

↓ ↑

Intelligence Layer

100 ITS Agents • LLM Gateway • Orchestrator Engine

↓ ↑

Data Layer

Google Cloud Storage • xAPI LRS • Cookieless Auth

Node.js 18+ Express Claude/GPT GCS xAPI Cloud Run

Speaker Notes

Key principles: No local database, stateless for Cloud Run, cookieless auth for GDPR.

🌍 Internationalization (i18n)

17 Languages • RTL Support • App-Wide

🇺🇸 English
🇨🇳 中文
🇪🇸 Español
🇩🇪 Deutsch
🇫🇷 Français
🇮🇹 Italiano
🇵🇹 Português
🇯🇵 日本語
🇰🇷 한국어
🇸🇦 العربية
🇮🇱 עברית
🇮🇷 فارسی
🇹🇭 ไทย
🇻🇳 Tiếng Việt
🇷🇺 Русский
🇮🇳 हिन्दी
🇵🇰 اردو

Usage: ?lang=zh or ?lng=es

RTL Languages: Arabic, Hebrew, Persian, Urdu automatically get dir="rtl"

LLM Integration: Language instruction injected into ALL prompts

Speaker Notes

When ?lang is set, the ENTIRE app must be in that language - no exceptions.

Performance & Scalability

94-96% Token Cost Reduction Through MD5 cache deduplication
<500ms Cached Response GCS content delivery
<10s New LLM Generation First-time content creation
100+ Concurrent Users Per Cloud Run instance
>90% Cache Hit Rate For repeated content
1000+ Events/Second Agent event processing

☁️ Stateless architecture enables unlimited horizontal scaling on Cloud Run

Speaker Notes

Token efficiency achieved through intelligent caching. Same competency + same config = same cache key.

📚 Research Foundation

Built on Decades of Learning Sciences

🧠 Cognitive Science

  • Cognitive Load Theory
  • Multimedia Learning
  • Working Memory Models
  • Schema Theory

🤖 ITS Research

  • GIFT Framework
  • Cognitive Tutors
  • BKT/PFA/DKT
  • Affect Detection

📊 Psychometrics

  • Item Response Theory
  • Computerized Adaptive Testing
  • Reliability & Validity
  • Knowledge Tracing

🎓 Pedagogical Methods

  • Socratic Method
  • Cognitive Apprenticeship
  • Reciprocal Teaching
  • Self-Determination Theory

Speaker Notes

Every feature in UALS is grounded in peer-reviewed research.

Speaker Notes

All links open in new windows. Feel free to explore during Q&A.

🏆 UALS vs. Alternatives

Traditional LMS (Moodle, Canvas)
UALS: 100% AI-generated content, 3,375 combinations vs static pre-authored
Adaptive Platforms (ALEKS, Knewton)
UALS: 100 agents + multi-philosophy vs single algorithm
AI Tutors (Khanmigo, etc.)
UALS: Complete ITS architecture vs single-agent chatbot
Competency Platforms
UALS: BOTH curriculum AND competency philosophies

Speaker Notes

Emphasize UALS is not just another chatbot - it's a complete intelligent tutoring system.

🔮 Future Roadmap

Phase 2 (Year 2)

+10 agents: Transfer Learning, Memory Consolidation, Multimodal Orchestration, Essay Scoring, Code Review...

Phase 3 (Year 3)

+10 agents: Learning Style, Cognitive Load, AR/VR, Voice Interaction, UDL...

Phase 4 (Year 4)

+7 agents: Attention Management, Game-Based Learning, Haptic Feedback, Stealth Assessment...

Ongoing

Additional philosophies (Montessori, constructivist), domain-specific agents, LTI/SIS integration, research collaborations

Speaker Notes

The architecture supports N philosophies and unlimited agent additions.

Questions & Discussion

📚 Documentation: /documentations/
🔬 Research collaborations welcome
💬 Live demo available

Thank you for your attention!

UALS - Universal Adaptive Learning System

Speaker Notes

Open for questions. Common topics: gaming detection, content accuracy, implementation timeline, cost, data privacy.