Technical Architecture · Learning Catalyst

How Learning Catalyst
Generates Courses in Minutes

A deep-dive into the AI pipeline, instructional design framework, and engineering decisions behind an e-learning authoring platform that reduces course production time by 99.5%.

i

About This Platform

This public version of Learning Catalyst demonstrates the platform's core architecture and capabilities. A production deployment of a similar system is used by over 160 registered users within one of the world's largest technology companies. Impact metrics shown below are based on that internal deployment and are provided as reference benchmarks — not yet measured on this public platform.

Impact

Measurable Outcomes

Reference metrics from the internal production deployment at a large technology company.

~3 min
Per course
vs 3–5 days traditional
160+
Registered users
Internal deployment
200+
Courses generated
Internal deployment
16,000+
Hours saved
Estimated total
£960k+
Cost saving
At £60/hr ID rate
99.5%
Time reduction
Per course produced
Architecture

System Architecture

A full-stack React + tRPC + Node.js application with a parallel AI orchestration layer. The frontend and backend share end-to-end TypeScript types via tRPC, eliminating the DTO mapping layer entirely.

System Architecture — Learning Catalyst
AI Pipeline

Parallel Generation Pipeline

Five AI services orchestrated in parallel to produce a complete course in under 4 minutes. The key innovation is firing the hero image simultaneously with the LLM content call, then parallelising all inline images and the Mermaid diagram.

1

Source Extraction

0–30s
  • PDF/DOCX/PPTX parsed server-side
  • URL scraping with YouTube transcript extraction
  • Multi-URL batch extraction (up to 10 URLs)
  • Normalised to a single extractedText string
2

Course Structure

30–60s
  • LLM generates module outline and lesson titles
  • Learning objectives aligned to Bloom's Taxonomy verbs
  • Audience and job family context injected into prompt
  • Validated JSON object drives all subsequent phases
3

Parallel Lesson Generation

60–180s
  • Batches of 3 lessons processed concurrently
  • LLM content + hero image fire simultaneously
  • 4 inline images + Mermaid diagram fire in parallel after blocks saved
  • json_object format with prompt-driven schema (not json_schema strict)
4

Assessment + Flashcards

180–210s
  • Summative quiz with Bloom's level tagging per question
  • Flashcard deck optimised for spaced repetition
  • Both generated after lesson content is complete
5

SCORM Packaging

210–240s
  • SCORM 2004 manifest (imsmanifest.xml) serialised
  • Each lesson rendered as self-contained HTML page
  • Package zipped and uploaded to S3
  • Signed download URL returned to client
Generation Sequence Diagram
Instructional Design

Learning Science Framework

Every lesson follows Gagné's Nine Events of Instruction, with Bloom's Taxonomy for cognitive depth and Mayer's Multimedia Learning Principles for visual design.

Gagné EventBlock TypeImplementationBloom's Level
1. Gain attentionText (hook)Provocative headline + 2–3 sentence opening
2. Inform objectivesLearning objectivesAuto-generated Bloom's verb-aligned objectivesAll levels
3. Stimulate recallCalloutColour-coded key insight card (tip/warning/info)Remember
4. Present contentFlip cards + TabsActive recall cards + multi-perspective tabbed contentUnderstand
5. Provide guidanceProcess stepperHorizontal stepper with numbered, colour-coded stepsApply
6. Elicit performanceChecklistActionable takeaway checklistApply
7. Provide feedbackQuiz (3 questions)Mixed Bloom's levels with answer explanationsAnalyse
8. Assess performanceSummative quizModule-level assessment with score trackingEvaluate
9. Enhance retentionFlashcard deckSpaced-repetition deck auto-generated from lessonRemember

Mayer's Multimedia Learning Principles are applied at the rendering layer: each AI image is positioned immediately adjacent to its related text block (spatial contiguity), and the hero image uses a gradient caption overlay rather than a separate caption element (temporal contiguity).

Tech Stack

Technology Decisions

Every technology choice was made to maximise developer velocity, type safety, and deployment simplicity on a serverless runtime.

Frontend
React 19 + Vite 6
Styling
Tailwind CSS 4
Components
shadcn/ui
API Layer
tRPC 11
Serialisation
Superjson
Backend
Express 4
Runtime
Node.js 22
Database
MySQL / TiDB
ORM
Drizzle ORM
AI — Text
LLM (GPT-4o class)
AI — Vision
Image Generation
AI — Audio
Whisper
Diagrams
Mermaid.js 11
AI — Video
D-ID API
Output
SCORM 2004

json_object over json_schema

The initial implementation used json_schema with strict: true and additionalProperties: false. This caused the model to return { type: 'text' } with no content — it interpreted the flat schema as requiring every field on every block type. Switching to json_object with a prompt-driven schema produced reliable, fully-populated output.

tRPC over REST + OpenAPI

tRPC eliminates the DTO mapping layer entirely. Drizzle ORM rows are returned directly from procedures, with Superjson handling Date serialisation. This removes an entire class of type-mismatch bugs and reduces the codebase by ~30% compared to an equivalent REST implementation.

Parallel image generation

The original sequential pipeline took ~66 seconds per lesson. Firing the hero image simultaneously with the LLM content call, then firing all 4 inline images and the Mermaid diagram in parallel after blocks are saved, reduced per-lesson time to ~22 seconds — a 3× improvement with no change to output quality.

Multi-Agent Pipeline

Six-Agent Orchestration

Rather than a single LLM prompt, Learning Catalyst decomposes course creation into six specialised agents. Each has a single responsibility, communicates via typed interfaces, and can be independently tested and improved.

1
Content ArchitectSource analysis, learning objective generation, course structure design
Bloom's Taxonomy5 Moments of NeedMerrill's First Principles
CourseOutline: { title, objectives[], modules[], contentStrategy }
2
Lesson WriterFull module content generation with instructional design rigour
Gagné's 9 EventsMayer's MultimediaKnowles' Andragogy
ModuleContent[]: { lessons[], interactiveElements[], pullItTogether }
3
Assessment DesignerPsychometrically valid formative + summative assessment generation
Assessment PsychometricsBloom's TaxonomyItem Writing Rules
AssessmentItems[]: { question, options[], correctAnswer, bloomsLevel, rationale }
4
Scenario BuilderBranching scenario generation with consequence modelling
Situated LearningProblem-Based LearningMerrill's Application
ScenarioTree[]: { situation, choices[], consequences[], optimalPath }
5
QA ReviewerAutomated quality evaluation against 12-criterion weighted rubric
Psychometric TheorySource Grounding Validation
QAReport: { overallScore, grade, criteria[], findings[], fixInstructions[] }
6
PackagerStandards-compliant SCORM/xAPI package assembly
SCORM 1.2SCORM 2004 3rd Ed.xAPI 1.0.3
PackageFile: { zipBuffer, manifest, scoStructure, launchUrl }
Assessment

Psychometrics Framework

Every assessment item is governed by 5 Core Values ensuring validity, reliability, fairness, transparency, and defensibility — mirroring professional test development standards.

1
Validity
Measures what it claims to measure
2
Reliability
Consistent results across administrations
3
Fairness
No demographic or cultural bias
4
Transparency
Clear criteria and rationale
5
Defensibility
Withstands independent scrutiny
Item Writing RuleDescription
Single correct answerOnly one option is unambiguously correct — no 'all of the above'
Clear stemQuestion is complete and understandable without reading the options
No negative phrasingAvoid 'Which is NOT...' or 'All EXCEPT...'
Homogeneous optionsAll options are the same type, length, and grammatical form
No absolute termsAvoid 'always', 'never', 'all', 'none'
Plausible distractorsWrong answers must be believable to someone who hasn't learned the material
Quality Assurance

12-Criterion QA Rubric

Every generated course is automatically evaluated by Agent 5 (QA Reviewer) against a weighted rubric. Courses scoring below 70 trigger an auto-fix pipeline before delivery.

#CategoryCriterionWeightPass
1ContentSource Content Coverage9/10≥70
2ContentFactual Accuracy10/10≥70
3ContentProgressive Structure (Gagné)8/10≥70
4ContentPerformance Objectives Quality8/10≥70
5AssessmentAssessment Validity (Core Value 1)9/10≥70
6AssessmentScenario-Based Question Quality8/10≥70
7AssessmentItem Writing Quality7/10≥70
8AssessmentBlueprint Alignment & Coverage7/10≥70
9IDLearner Engagement6/10≥70
10IDAudience Appropriateness5/10≥70
11AccessibilityFairness & Bias (Core Value 5)5/10≥70
12AccessibilityClarity & Readability4/10≥70
A
90–100
Pass — deliver
B
80–89
Pass — minor notes
C
70–79
Pass — minimum
D
60–69
Auto-fix triggered
F
0–59
Critical auto-fix
Hallucination Prevention

Source Grounding Validation

A programmatic (non-LLM) validation runs as part of the QA pipeline to independently verify that generated content is grounded in the source material.

Algorithm

  1. 1Extract key terms from source (capitalised phrases, ALL-CAPS, quoted terms)
  2. 2Filter generic terms ('introduction', 'overview', 'module')
  3. 3Rank by specificity — longer terms are more distinctive
  4. 4Sample top 30 terms for coverage check
  5. 5Case-insensitive substring match across all generated output
  6. 6Compute coverage percentage
Blended Scoring
40% weight from the LLM's content coverage evaluation + 60% weight from the programmatic key-term check. This prevents the LLM from "grading its own homework."
Coverage Thresholds
≥ 70%Pass
40–69%Flag for review
< 40%Auto-fix triggered
Multilingual Support
Content generation in 40+ languages. The targetLanguage parameter flows from CourseWizard → tRPC procedure → LLM system prompt → all 7 block types → image prompt translation → SCORM manifest xml:lang attribute.
Security

Security Architecture

Defence-in-depth across authentication, authorisation, input validation, and storage.

LayerMechanismImplementation
AuthenticationOAuth 2.0 (SSO)JWT session tokens; secure httpOnly cookies
AuthorisationRole-based access control (RBAC)super_admin, admin, user roles with graduated permissions
Input SanitisationZod schema validation on all inputsPrevents injection; enforces type safety at runtime
Data IsolationPer-user project scopingUsers can only access their own courses
API SecuritytRPC protected proceduresAuth context injected automatically; no manual checks needed
Storage SecurityS3 with scoped accessFiles accessible only via signed URLs
Transport SecurityHTTPS everywhereTLS 1.3; HSTS headers
Secrets ManagementEnvironment variable injectionNo hardcoded credentials; platform-managed secrets
Resilience

Error Handling & Resilience

The multi-agent pipeline is designed to degrade gracefully. No single agent failure should prevent course delivery.

PatternTriggerRecovery
Circuit Breaker3 consecutive agent failuresSkip agent; use fallback output
Exponential BackoffTransient API errors (429, 503)Wait 1s → 2s → 4s, then fail gracefully
Graceful DegradationContent Architect failureFall back to single-prompt generation
Timeout ProtectionAgent exceeds 60sTerminate and use partial output
Output ValidationMalformed agent outputRetry once, then use fallback
Idempotent RetriesAny transient failureSame input → same structural output; safe to retry
Engineering Quality

Quality Metrics

100% TypeScript strict mode across the entire codebase. Tests cover unit, integration, contract, and resilience scenarios.

905
Automated Tests
Unit + Integration + Contract
51,424
Lines of Code
TypeScript strict mode
129
Git Commits
Consistent cadence
100%
Type Coverage
Zero any types
Test TypeCountWhat It Validates
Unit Tests~600Individual functions, utilities, schema validation
Integration Tests~200Agent pipelines, database operations, API endpoints
Contract Tests~80Agent input/output type conformance
Resilience Tests~25Circuit breakers, timeouts, fallback behaviour
Recognition

2026 Executive Review Selection

A concise programme-selection record, presented separately from product, performance, and affiliation claims.

Internal programme record

Shortlisted for executive review

In a 2026 global innovation programme at a major technology company, Learning Catalyst was selected as one of six tools for executive review from 211 projects created by approximately 600 participants. Senior technology leadership attended the review.

This statement is based on the project creator's internal programme record and describes programme selection and review attendance only. It does not state or imply corporate endorsement, sponsorship, partnership, or affiliation.

Roadmap

What's Built vs What's Next

v1.0 is live and fully functional. v2.0 focuses on adaptive learning, analytics, and collaboration features.

Completed (v1.0)

Multi-agent orchestration (6 agents)
Psychometric assessment generation
QA Agent with auto-fix pipeline
SCORM 1.2/2004 + xAPI export
Source grounding validation
Branching scenarios
AI image generation (5 per lesson, parallel)
Mermaid flowchart per lesson
AI presenter video (D-ID)
AI Tutor Chat
Blended learning blocks (7 types)
Flashcard decks
My Courses (localStorage persistence)
Multilingual content generation (40+ languages)

Planned (v2.0 — H2 2026)

Multi-language UI (full RTL support)High
Adaptive learning pathwaysHigh
Assessment analytics dashboardMedium
SME review workflowMedium
Template marketplaceMedium
xAPI analytics integrationLow
Content Blocks

14 Visual Block Types

Each block type is a distinct React component with its own visual design, interaction model, and instructional purpose.

T
Text
Editorial prose with heading
!
Callout
Colour-coded insight card
Flip Cards
Active recall interaction
Tabs
Multi-perspective content
Process List
Horizontal stepper
List
Checklist / numbered / bullet
?
Quiz
Formative assessment
AI Image
Contextual generated image
Mermaid
Auto-generated flowchart
Timeline
Milestone visual
Quote
Pull quote with attribution
Accordion
Collapsible sections
Video
Embedded video player
Objectives
Bloom's-aligned goals

See It in Action

Upload a PDF, paste a URL, or type a topic — get a complete, SCORM-compliant course with AI images, interactive blocks, and a quiz in under 4 minutes. No login required.