A high-stakes IP certification provider · Knowledge architecture & certification content engineering
Digitizing an expert's knowledge for certification 500,000 words, one system.
A fragmented body of expert material (116 documents, roughly 500,000 words) extracted into a RAG knowledge base and rebuilt as a full certification ecosystem across self-paced, virtual, and classroom formats.
participant satisfaction score
source documents extracted and indexed
topics across 5 domains
exam-rigor assessment questions engineered
The certification's substance lived in one place: its experts. The source material, 116 documents, roughly 500,000 words of slide decks, study guides, and raw notes - was fragmented, inconsistent, and unusable as a curriculum. The mandate was to digitize that expertise into a production-grade certification ecosystem spanning self-paced modules, virtual instructor-led sessions, and classroom delivery, with an assessment bank engineered to genuine exam rigor.
The domain made it harder. IP enforcement, trade-secret protection, counterfeit-investigation logic - specialized material that can't be interpreted by generalist production staff, validated by experts whose availability was scarce and whose feedback often identified that something was wrong without specifying what right looked like.
First, the knowledge architecture. All 116 documents were extracted, indexed, and tagged into a retrieval-augmented generation (RAG) database (bibliographies, metadata, more than 20 structured columns) with an LLM chatbot front end for summary and information-level querying. That infrastructure fed AI-assisted rewrite workflows for the curriculum itself.
On top of it: 42 topics across five domains, delivered as self-paced modules, six virtual instructor-led sessions, and full classroom materials with facilitator guides. When early drafts proved too text-dense to teach from, the whole storyboard process pivoted to visual-first (icons, flowcharts, and diagrams replacing prose) because for dense technical material, experts commit reliably only to what they can see.
The assessment bank was its own engineering problem: more than 200 discrete-option multiple-choice questions built to exam standard. When early questions missed the rigor bar, we rebuilt the process around a collaborative style guide derived from the expert's own edits, turning corrective feedback into a reusable specification.
“An expert's knowledge isn't a curriculum until someone architects it.”
By mid-2025 the full ecosystem had shipped: the indexed knowledge base, 42 topics across three delivery modalities, classroom materials, and the 200-plus question bank. The delivered modules scored 4.91 out of 5 with participants, on a certification where exam rigor, not learner comfort, sets the bar.
What made it hard
The feedback physics. At peak, topics ran through 5 to 10 review rounds, and the production workflow was rebuilt six times to absorb shifting expert requirements. Final approval rested entirely on two experts with no backup reviewer, so the project stalled outright whenever either was unavailable. And because the domain was too specialized for general production staff to interpret, senior team members hand-built the flowcharts, wiring diagrams, and logic references that production needed, knowledge engineering absorbed inside an instructional-design engagement.