← Back to work

Flagship case study

Architecting an audit-defensible safety certification program

Client: a confidential defense / advanced-manufacturing client · Role: lead instructional designer & de facto program manager · Status: in active build

01 — The mandateCertify a workforce on equipment that can kill — and make it survive an audit

The client had to certify employees across many areas of workplace safety, at scale, with turnover, and across multiple work areas. The catch: under OSHA, an operator certification isn’t real unless it’s defensible — formal instruction, hands-on practical evaluation, and a documented record that holds up when an inspector asks for it. A pile of “completed” e-learning doesn’t satisfy that. The brief was effectively: build a certification system, not a course library.

02 — The insightDon’t build N courses. Build one core that every credential inherits.

Most of the safety content — hazard awareness, required pre-shift checks, and site rules — is identical across roles. So instead of authoring redundant courses per role, I designed a reusable shared-core spine built once and inherited by every credential, with only the role-specific differences layered on top. One source of truth, consistent everywhere, and updates propagate to every program at once.

03 — The systemA four-layer architecture from objective to certification record

  • Shared core (async): the cross-cutting safety foundation, authored once.
  • Role-specific (async): the procedures, checkpoints, and hazards unique to each role.
  • Function modules: role-specific add-ons (e.g., specialized procedures) where the job demands them.
  • Standardized in-person practicum: a facilitator-guided practical evaluation that produces the legally required certification record.

Every program is anchored to its specific regulatory floor (the relevant OSHA and ANSI standards), so each design decision traces back to a requirement an auditor can check.

04 — The engineeringI built it, not just designed it

This is where the program differs from a typical instructional-design engagement. I didn’t storyboard and hand off:

  • Hand-coded interactions — self-contained, accessible HTML/CSS/JS web objects (card sorts, audit simulators, decision flows) with full ARIA and keyboard support, built on a versioned, tokenized design system.
  • LMS integration spec — a SCORM/xAPI completion model plus a certification state machine that enforces the rule the whole program depends on: completion does not equal certification until the practical evaluation is passed.
  • AI-assisted production pipeline — scripted voiceover, captioning, and imagery, all governed end-to-end by a human accessibility/QA gate.

05 — The governanceThe system that keeps the system shipping

A multi-program build only survives if scope and quality are controlled. I authored the governance layer myself: a single-funnel SME-review playbook with Need-vs-Nice-to-Have adjudication, a RACI matrix, go/no-go gates, and weekly leadership status reporting — plus a self-authored WCAG 2.1 AA implementation guide (hand-calculated contrast, ARIA guidance, and a documented testing toolchain) so accessibility is a standard, not an afterthought.

06 — The proofHonest evidence the method works

The certification program is in active build, so its own outcome data isn’t in yet. But the same design method — applied to a parallel 10-course ISO 9001 compliance curriculum I built for the same client — produced measured results: a +1.3-point average gain on a 5-point scale, improvement in 100% of measured courses, and a 4.8/5 average rating across 200+ reviews and 3,000+ survey responses. (These are self-reported knowledge/confidence gains, not graded-exam scores — measured with identical pre/post instruments.)

A note on confidentiality

The client, the industry specifics, and identifying details are withheld by design — some of my best work is for organizations I can’t name, and discretion is part of the job. Full technical detail (the architecture documents, the LMS integration spec, and sample interactions) is available under NDA on request.

Discuss this work