Overview
UMCF (UnaMentis Curriculum Format) is a JSON-based specification for educational content optimized for voice-based AI tutoring, built from the ground up for natural conversation.
Key Characteristics
- Voice-Native: Every text field can have TTS-optimized spoken variants
- Tutoring-First: Built for natural conversation, not LMS delivery
- Unlimited Hierarchy: Topics can nest to arbitrary depth
- Standards-Grounded: Maps to IEEE LOM, LRMI, SCORM, xAPI, QTI, CASE, Open Badges
- AI-Enrichment Ready: Designed for automated content enhancement
Here's an example of the basic UMCF document structure in JSON:
{
"umlcf_version": "1.0.0",
"curriculum": {
"id": "intro-calculus",
"title": "Introduction to Calculus",
"description": "A comprehensive introduction to differential calculus",
"language": "en-US",
"topics": [...]
}
}
What UMCF Provides
UMCF is designed specifically for the unique needs of voice-based AI tutoring:
| Capability | UMCF Feature |
|---|---|
| Pronunciation guides | Built-in spokenText variants |
| Comprehension checks | Inline stopping points |
| Alternative explanations | First-class support |
| Misconception handling | Remediation triggers |
| Conversation flow | Natural dialogue structure |
| Depth adaptation | Multiple depth levels |
Document Structure
A UMCF document has a hierarchical structure:
Voice-Native Features
UMCF includes several features specifically designed for text-to-speech synthesis and natural voice interaction.
Spoken Text Variants
Every text field can have a spokenText variant optimized for TTS:
{
"text": "The derivative of x² is 2x.",
"spokenText": "The derivative of x squared is 2 x."
}
Pronunciation Guides
The glossary supports pronunciation hints for technical terms:
{
"glossary": [{
"term": "Euler",
"definition": "Swiss mathematician (1707-1783)",
"pronunciation": "OY-ler",
"ipa": "/ˈɔɪlər/"
}]
}
Stopping Points
Inline comprehension checks that pause the AI to verify understanding:
{
"transcript": {
"paragraphs": [
"A derivative measures the rate of change of a function.",
"Think of it as the slope of the function at any point."
],
"stoppingPoints": [{
"afterParagraph": 1,
"type": "comprehensionCheck",
"prompt": "Can you think of a real-world example where we care about rate of change?"
}]
}
}
Content Depth Levels
UMCF supports multiple depth levels for the same content, allowing the AI tutor to adapt to the learner's needs:
High-level intuition without technical detail. Perfect for quick introductions.
Basic concepts with minimal mathematics. Suitable for beginners.
Moderate detail with equations mentioned but not derived.
In-depth coverage with derivations and proofs.
Comprehensive treatment with full mathematical rigor.
Paper-level depth with citations and cutting-edge content.
Topic Structure
Each topic contains the content and metadata for a single learning unit. Here's an example topic definition in JSON:
{
"id": "derivatives-introduction",
"title": "Introduction to Derivatives",
"depthLevel": "intermediate",
"estimatedDuration": "PT20M",
"learningObjectives": [
"Define the derivative as a rate of change",
"Calculate derivatives of polynomial functions",
"Interpret the derivative geometrically as slope"
],
"transcript": {
"text": "Today we're going to explore derivatives...",
"spokenText": "Today we're going to explore derivatives.",
"stoppingPoints": [...]
},
"alternativeExplanations": [{
"id": "simple-analogy",
"style": "analogy",
"content": "Think of it like a speedometer in your car..."
}, {
"id": "technical",
"style": "technical",
"content": "Formally, the derivative is defined as the limit..."
}],
"misconceptions": [{
"id": "derivative-is-tangent",
"description": "Confusing derivative with tangent line",
"triggers": ["the derivative is a line", "derivative touches the curve"],
"remediation": "The derivative gives us the slope of the tangent line..."
}],
"children": [
{ "id": "power-rule", ... },
{ "id": "chain-rule", ... }
]
}
Alternative Explanations
When a learner doesn't understand, the tutor can try a different approach:
- simple: Plain language, no jargon
- technical: Formal definitions and notation
- analogy: Real-world comparisons
- visual: Descriptions of diagrams or animations
- example: Worked examples
Misconception Handling
UMCF allows content authors to anticipate common misunderstandings:
- Triggers: Phrases that indicate the misconception
- Remediation: Corrective explanation
- Prerequisites: Topics to review if misconception persists
Assessments
UMCF supports various assessment types optimized for voice interaction:
Reflection Questions
Open-ended prompts for deeper thinking
{
"type": "reflection",
"prompt": "How might derivatives help
in predicting stock prices?",
"rubric": [
"Mentions rate of change",
"Connects to real-world prediction"
]
}
Verbal Quizzes
Quick knowledge checks with spoken answers
{
"type": "verbalQuiz",
"question": "What is the derivative
of x cubed?",
"expectedPatterns": [
"3 x squared",
"three x to the second"
],
"feedback": {
"correct": "Exactly right!",
"incorrect": "Remember the power rule..."
}
}
Application Problems
Real-world problem solving scenarios
{
"type": "application",
"scenario": "A ball is thrown upward...",
"questions": [
"When is the ball at its highest?",
"What is the maximum height?"
],
"hints": [
"What happens to velocity at the peak?"
]
}
Learning Science Features
UMCF includes pedagogical features grounded in learning science research to promote genuine understanding and long-term retention.
Teachback Checkpoints
Students explain concepts in their own words. The AI evaluates depth and accuracy with tiered feedback.
{
"type": "teachback",
"prompt": "In your own words, explain
what a derivative represents.",
"tiers": {
"excellent": "≥0.9 accuracy",
"good": "≥0.7 accuracy",
"partial": "≥0.4 accuracy",
"struggling": "<0.4 accuracy"
}
}
Spaced Retrieval
Key concepts are flagged for periodic review using proven spacing algorithms to strengthen memory.
{
"retrieval": {
"algorithm": "sm2",
"intervals": [1, 3, 7, 14, 30],
"conceptId": "derivative-definition",
"prompt": "What does a derivative
measure?"
}
}
Productive Struggle Metrics
Tracks and celebrates cognitive effort. Think time before responding is measured and encouraged.
{
"metrics": {
"thinkTimeSeconds": 45,
"teachbackAttempts": 2,
"encouragement": "You invested
45 seconds of deep thinking.
That's how real learning happens."
}
}
Import System
UMCF includes a pluggable import system for converting existing content:
MIT OpenCourseWare
Import courses from MIT OpenCourseWare
OCW → UMCF
Complete
CK-12 FlexBooks
Import K-12 EPUB textbooks from CK-12 Foundation
EPUB → UMCF
Complete
EngageNY
Import K-12 curriculum from EngageNY
EngageNY → UMCF
Complete
MERLOT
Import learning materials from MERLOT repository
MERLOT → UMCF
Complete
Fast.ai Notebooks
Import Jupyter notebooks from fast.ai courses
Jupyter → UMCF
Spec Complete
Stanford SEE
Import courses from Stanford Engineering Everywhere
SEE → UMCF
Spec Complete
AI Enrichment Pipeline
Raw imported content can be enhanced using a 7-stage LLM pipeline:
- Content Analysis: Identify structure and key concepts
- Segmentation: Break into conversational chunks
- Objective Extraction: Generate learning objectives
- Assessment Generation: Create quizzes and reflections
- Tutoring Enhancement: Add stopping points and misconceptions
- Alternative Explanations: Generate multiple explanation styles
- Knowledge Graph: Build concept relationships
AI Curriculum Generation
In addition to importing existing content, UnaMentis provides a comprehensive prompt system for generating UMCF-compliant curriculum from scratch using AI models like Claude or GPT-4.
Generation Prompt Features
- Complete Format Compliance: Generates valid UMCF v1.1.0 JSON
- Voice-Optimized Content: Speaking notes, pronunciations, and TTS-friendly text
- Teachback Checkpoints: Comprehension verification built in
- Bloom's Taxonomy Alignment: Learning objectives at appropriate cognitive levels
- Assessment Generation: Quizzes with feedback included
- Misconception Handling: Common errors with remediation strategies
How to Use
Copy the Prompt
Get the generation prompt from the documentation.
Add Your Specification
Specify the topic, target audience, scope, and desired depth level.
Generate with AI
Submit to Claude, GPT-4, or another capable model.
Review and Save
Validate the output JSON and save with a .umcf extension.
This approach lets you create custom curriculum for any topic, tailored to your specific audience and learning goals, without requiring technical knowledge of the UMCF format.
Standards Mapping
UMCF fields map to established educational technology standards for interoperability:
| Standard | Purpose | UMCF Mapping |
|---|---|---|
| IEEE LOM | Metadata | curriculum.metadata.* |
| LRMI | Learning resources | topic.learningObjectives, topic.depthLevel |
| SCORM | Content packaging | curriculum structure, sequencing |
| xAPI | Learning analytics | assessment events, progress tracking |
| QTI | Assessments | topic.assessments.* |
| CASE | Competency frameworks | learningObjectives, masteryThresholds |
| Open Badges | Credentials | curriculum.badges, topic.achievements |
Specification Resources
- JSON Schema: 1,905 lines (Draft 2020-12)
- Human-Readable Spec: Complete documentation
- Example Curricula: Minimal and realistic examples
- Validation Tools: Schema validators and linters