Learning Architecture

I build systems that let other people learn, decide, and deliver without me in the room.

Learning architecture, for me, is not content production. It is the design of structures, interfaces, assessments, and coaching loops that make high-quality performance repeatable. This page shows that practice across enterprise enablement, adaptive coaching, and a modular AI-native curriculum system.

20
Years Designing Learning
121
Curriculum Modules
3
Assessment Layers
12K+
Users / Month
Ask Pathfinder

Turning scattered expertise into a system people can actually use at the moment they need it.

From Facilitated Knowledge to Self-Serve Performance
1
Source Layer
Field content, examples, updates, and raw knowledge spread across teams.
presentationsdocsexamplespartner feedback
▼ structure
2
Semantic Layer
Function-based taxonomy, parent-child relationships, and human-curated metadata.
taxonomymetadatafreshness rulescompetency signals
▼ usable at scale
3
Learning Layer
Better discovery, reusable templates, and scalable self-serve enablement.
12K+ users/mo25% faster prep67% better coverage
EnablementKnowledge ArchitectureMetadataScale

Ask Pathfinder — High-Touch Expertise Into Scalable Discovery

At AWS, I helped transform high-touch sales enablement into a structured learning and discovery system. The core move was architectural, not cosmetic: reclassify content by function instead of industry, add parent-child content relationships, and layer human-designed metadata over source materials so people could find what mattered quickly and trust what they found.

What Changed

  • Industry-based organization became function-based classification
  • Content relationships became explicit instead of implied
  • Metadata carried human judgment, not just automated tags

Why It Matters

  • People found relevant guidance faster
  • Knowledge transferred across use cases instead of staying siloed
  • The system scaled without depending on live facilitation
The SDK for People

A modular learning system built to help non-experts build real AI capability through artifacts, not slides.

Orient
Build
Verify
121 modules
5 journeys
Artifact-based
RIU-401 Taxonomy Design
structure a domain for retrieval
RIU-021 Eval Harness
measure model quality with evidence
RIU-510 Workflow Design
route work across agents safely
RIU-102 Enablement Pack
teach the next person clearly
CurriculumAssessmentAI-NativeBuilder of Builders

Enablement System — The SDK for People

This system turns a broad AI capability map into structured learning journeys, hands-on modules, and visible progression. Each module is tied to a competency, each competency produces a real artifact, and each artifact can be evaluated by automated checks, AI rubric scoring, and human calibration. The result is a learning architecture that behaves more like a product system than a course catalog.

System Design

  • 121 modules mapped to explicit competency areas
  • 5 learning journeys instead of one generic track
  • Knowledge library and taxonomy share the same backbone

Assessment

  • Automated checks verify structure and completeness
  • AI rubric scoring evaluates reasoning against exemplars
  • Human calibration handles edge cases and final trust

Why It Works

  • The learner makes real things, not quiz selections
  • Progress is visible through artifacts
  • Quality scales without flattening judgment
Adaptive Coaching Loop

Learning systems should respond to confusion, confidence, and context, not just deliver the next lesson.

Oka
Adaptive learning companion
Oka says
You do not need to know the jargon yet. Let’s start with the part you can already name.
State: orienting
next step
profile → converge → try one artifact → verify → adapt
Talk
CoachingAdaptive UXPlain LanguageGovernance

Adaptive Coaching — A Knowledge Companion You Can Ask

This is a conversational companion built on top of a structured knowledge base — like Ask Pathfinder, but for any domain. You ask it how to build something, and it routes your question through a taxonomy of 121 modules, retrieves the relevant knowledge entries, and walks you through the answer with citations. It adapts to your level: if you're stuck, it simplifies. If you're skipping ahead, it fills gaps. If you're overwhelmed, it narrows scope. The coaching loop responds to real learner states, not a fixed curriculum sequence.

How It Works

  • You ask a question in plain language
  • The system classifies it against a 121-node taxonomy
  • Retrieves sourced knowledge entries (176 entries, 565 citations)
  • Responds with a guided answer + what to try next

Adaptive Behavior

  • Stuck: simplifies, offers a smaller starting point
  • Skipping ahead: fills prerequisite gaps before advancing
  • Overwhelmed: narrows to one concrete next step
  • Ready: advances to the next module with an artifact to build

Design Principles

  • Glass-box: the learner can see why the system recommended something
  • No jargon without translation
  • Works for executives, technical teams, and children
  • The system teaches through interaction, not content delivery
The path here
2006–2013
Paris teaching and curriculum design
Sciences Po, Nanterre, Ecole des Metiers du Livre
Designed immersive, domain-shaped learning experiences where language and professional practice were taught together.
2013–2023
Operational learning and knowledge systems
Amazon
Built training, classification, and multilingual knowledge structures that scaled across teams, countries, and operating contexts.
2024–2026
AI-native enablement systems
AWS and independent systems work
Moved from programs and content into AI-native learning architecture: retrieval, adaptive coaching, modular curriculum, and artifact-based assessment.
How I think about this

Learning architecture is operating design

The goal is not to explain things beautifully in isolation. The goal is to create structures, workflows, and interfaces that help people perform better when the work is real.

The SDK for people

A good learning system gives people reusable primitives: shared language, visible standards, guided sequences, and feedback loops they can use without the original expert present.

Artifacts beat abstractions

If the learner cannot produce something real, the system has not taught enough. Briefs, taxonomies, harnesses, workflows, and enablement packs are the assessment.

Adaptive does not mean opaque

A strong coaching system changes its behavior as the learner changes, but it remains glass-box: the learner can see why the system recommended the next step.

Want to see the system in action?

Explore the open-source architecture, review a learning path, or reach out to discuss learning systems for AI-native organizations.