Creating Interactive AI Content

Curator v1.3 OAS v1beta1 Gemini 2.0

The Curator Pipeline is ColabEdu’s AI engine that transforms educational documents into structured, assessable interactive content — ready to be consumed by the A2UI platform.

What can the Curator generate?

The Curator produces three main types of interactive specifications:

InteractiveLesson

Multimedia lessons with slides, interactive widgets, adaptive quizzes and STEM drawing canvases.

Activated with: --output-type INTERACTIVE_LESSON

ExerciseSpec

Assessable exercises with AI rubric: essays, quizzes, Socratic debate, digital whiteboards and guided activities.

Activated with: --output-type EXERCISE

ResourceLearning

Enriched pedagogical resources: flashcards, infographics, glossaries and curricular reference materials.

Activated with: --output-type RESOURCE_LEARNING

Supported input sources

The Curator accepts content from multiple origins:

The most powerful flow: transforms exported LMS packages directly into ColabEdu interactive lessons.

Format	Source Platform	Notes
`.imscc`	Canvas LMS	HTML pages, assets and XML manifest
`.mbz`	Moodle	Full backup with activities and resources
`.zip` SCORM	Any LMS	Structured HTML + asset map

Example CLI usage:

ce-cli.sh curator curate \
  --file ./biology_bach1.imscc \
  --output-type INTERACTIVE_LESSON \
  --adaptive \
  --locale en

The --adaptive flag activates dynamic quiz routing (on_success / on_fail), enabling personalized navigation paths based on student performance.

The Curator can process curricular documents in PDF or Word and automatically extract the pedagogical structure (modules → lessons → slides).

Supported formats:

PDF (with OCR for scanned documents)
DOCX / DOC
Structured Markdown

Example:

ce-cli.sh curator curate \
  --file ./biology_syllabus.pdf \
  --output-type INTERACTIVE_LESSON \
  --locale en \
  --curricular-ref global.c0.ib.spanish_b.hl.paper1.v1

For programmatic integrations, the Curator accepts content extracted directly from the Canvas or Moodle API.

REST API parameters:

POST /api/v1/curator/curate
{
  "inputVariables": {
    "lms_manifest_xml": "<manifest>...</manifest>",
    "extracted_html_content": "=== RESOURCE: Lesson 1 ===\n...",
    "asset_url_map": {
      "images/diagram.png": "https://storage.googleapis.com/..."
    },
    "with_quiz_routing": true
  },
  "outputType": "INTERACTIVE_LESSON",
  "locale": "en"
}

The Curator can operate in headless batch mode over a full ce-specs repository, ingesting multiple specifications in a single run.

# Sync and ingest the full local catalog
./sync_specs.sh local GLOBAL

# Sync from Git (production)
./sync_specs.sh git GLOBAL

See Catalog Reference for the full list of available IDs.

Available Exercise Types

When the Curator generates an ExerciseSpec, it selects the most appropriate ExerciseType based on the curricular context:

ExerciseType	Description	Evaluation
`core.etype.mixed_quiz.v1`	Mixed quiz (MCQ + fill-blank)	Deterministic AI
`core.etype.split_pane_essay.v1`	Guided essay in split-pane	AI Rubric
`core.etype.socratic_tutor.v1`	Socratic tutor (chatbot)	AI Log + rubric
`core.etype.socratic_debate.v1`	Oral Socratic debate	AI Log
`core.etype.multimodal_mcq.v1`	MCQ with image / chart	Deterministic AI
`core.etype.grouping.v1`	Drag-and-drop classification	Deterministic AI
`core.etype.matching.v1`	Matching pairs	Deterministic AI
`core.etype.ordering.v1`	Chronological ordering	Deterministic AI
`global.exercise_type.free_text.v1`	Free essay / open response	AI Rubric
`global.exercise_type.mcq.v1`	Classic multiple choice	Deterministic AI
`global.exercise_type.guided_activity.v1`	AI-guided activity	AI Log + rubric
`global.exercise_type.whiteboard.v1`	Digital whiteboard (free draw)	Vision AI (PNG)

Interactive Lesson — Slide Anatomy

An InteractiveLesson is a sequence of slides, each with a layout and optional widgets. The Curator acts as an autonomous Art Director: it selects the most appropriate layout for each content block.

Available layouts

Layout	Optimal use
`title`	Cover — large title + background image
`split`	Concept + interactive widget (50/50)
`quote`	Definition, axiom or law (≤60 words, dark background)
`image`	Full-screen image with caption
`full`	Mermaid diagram or full-screen widget
`content`	Long text in enriched Markdown
`video`	Video player with transcript

Interactive widgets

Widgets are embedded in any slide with the embeddedWidget syntax:

embeddedWidget:
  type: quiz_widget          # ← Comprehension check
  props:
    questions: [...]
    routes:                  # Adaptive routing (activated with --adaptive)
      on_success: slide_advanced
      on_fail: slide_review

Available widgets:

quiz_widget — Quiz with adaptive routing on_success / on_fail
mermaid_viewer — Diagram auto-generated by the Curator
markdown_editor — Open text editor (synthesis, micro-essay)
grouping_widget — Drag-and-drop classification
scratchpad_widget — Free-draw canvas for STEM
whiteboard_widget — Digital whiteboard with Gemini Vision evaluation

Digital Whiteboard New

The global.exercise_type.whiteboard.v1 ExerciseType enables drawing exercises automatically evaluated by Gemini Vision on the PNG captured from the student’s canvas.

STEM use cases

Mathematics

Equation solving, function graphing, geometry and linear algebra. The student shows the step-by-step process on the canvas.

Sciences

Force diagrams (Physics), chemical reactions, anatomical schematics (Biology) and electrical circuits.

Evaluation flow

Student draws on canvas
        ↓
[PNG captured on "Submit"]
        ↓
Backend → Gemini Vision API
        ↓
C0 Rubric applied (global.c0.whiteboard.stem.math.v1)
        ↓
Feedback on 4 criteria:
  · Completeness  · Conceptual correctness
  · Clarity       · Process visibility

Recipe configuration

apiVersion: colabedu.ai/v1beta1
kind: Recipe
metadata:
  id: global.c1.stem.math.functions.whiteboard.v1
spec:
  assembly:
    param_refs:
      - global.exercise_type.whiteboard.v1
    rubric_refs:
      - global.c0.whiteboard.stem.math.v1
  interaction_mode: split_pane
  evaluation:
    capture_format: png
    evaluator: vision_ai
    feedback_on_submit: true

Full flow: from document to lesson

LMS Document (.imscc / .mbz / .pdf)
              ↓
     [Curator CLI / REST API]
              ↓
     NativeMultimodalDocumentProcessor
              ↓
     Curator Gem (curator_lesson v1.3.0)
              ↓
     InteractiveLesson YAML → SpecManager
              ↓
     A2UI (Flutter) renders the lesson

CLI reference commands

# Interactive lesson from LMS package with adaptive routing
ce-cli.sh curator curate \
  --file ./canvas_export.imscc \
  --output-type INTERACTIVE_LESSON \
  --adaptive \
  --locale en

# STEM whiteboard exercise from math PDF
ce-cli.sh curator curate \
  --file ./math_calculus.pdf \
  --output-type EXERCISE \
  --exercise-type global.exercise_type.whiteboard.v1 \
  --locale en

# Test with predefined manifest
ce-cli.sh curator test \
  --manifest ./tests/lms_curator_test_manifest.yaml \
  --output-type INTERACTIVE_LESSON

OAS YAML Reference — Complete syntax for all spec types
Exercise Types (C1) — Full ExerciseType catalog
Catalog Reference — SpecManager canonical IDs
CLI Tutorial — Step-by-step CLI curation guide
Curator Capabilities — Curator pipeline architecture