I design learning systems
that solve real business problems.
Senior learning experience designer specializing in high-stakes operational training, AI-augmented design, and knowledge systems for global tech teams. I work end-to-end — from needs analysis and stakeholder strategy to blended program architecture and measurable outcomes.
5 projects. End to end.
Each project covers the full design lifecycle — from needs analysis and stakeholder strategy through final delivery, measurement, and reflection. Topics are drawn from real operational contexts at global tech companies.
Knowledge Base Architecture for a Trust & Safety Ops Team
Taxonomy, governance model, and SOP system designed so a 24/7 global escalation team can find the right answer in under 60 seconds during a live incident.
Overview
A global platform's 24/7 escalation team was managing YouTube's highest-priority brand reputation and user safety incidents — but their knowledge base had grown organically for three years without governance. Articles were duplicated, outdated, inconsistently structured, and buried under a flat tag system with no hierarchy.
During a live incident, escalation managers were spending 4–7 minutes finding the relevant SOP — time that directly affected response speed on brand-critical events. The team needed a knowledge system, not just a document folder.
My task was to design the architecture from the ground up: taxonomy, content standards, governance workflow, and SOP templates that any team member in any timezone could use without training on the system itself.
Process
I audited 140+ existing KB articles, categorizing by topic, recency, accuracy, and usage frequency. I identified 38% as outdated, 22% as duplicates, and 15% as unfindable due to poor tagging. This audit became the case for a full restructure rather than incremental patches.
I designed a three-tier taxonomy: domain (e.g., Brand Safety, Creator Escalations, Policy Violations), incident type (e.g., Coordinated Inauthentic Behavior, Copyright, Misinformation), and content type (SOP, Reference, FAQ, Escalation Path). I tested findability with 5 users using a card sort and think-aloud before finalizing.
I designed a modular SOP template with mandatory fields: trigger conditions, decision tree, escalation path, owner, last-reviewed date, and linked resources. The template structure allowed global teams to scan — not read — an SOP during a live incident. Target scan time: under 90 seconds.
I established a content lifecycle: quarterly review cycles assigned to named owners, a deprecation workflow for outdated content, a new article approval checklist, and a "helpfulness rating" feedback mechanism tied to post-incident reviews. The governance model was designed to run without an L&D team member as gatekeeper.
Deliverables
Results & Takeaways
The findability test after restructuring showed average SOP location time dropped from 4–7 minutes to under 90 seconds. The governance model gave team leads a self-maintaining system — content quality no longer depended on an L&D team member reviewing every update.
Knowledge management is an instructional design problem, not a content management problem. The architecture — how information is structured, named, and connected — determines whether people can learn from it under pressure. This is where ID skills translate directly into operational impact.
Using GenAI to Accelerate SOP and Training Content Development
A documented case study of using LLMs as a design accelerator — showing prompts, outputs, bias evaluation, and the human judgment layer that makes AI-assisted content trustworthy at scale.
Overview
Both Amazon and Google explicitly call for designers who can leverage generative and agentic AI for content creation, personalization, and rapid prototyping while critically evaluating outputs for accuracy, bias, and ethical standards. This project documents exactly that process — not AI as a topic, but AI as a tool in my design workflow.
The context: a global ops team needed 12 updated SOPs and 3 microlearning modules within a 6-week window that would normally require 14 weeks. I used GenAI to compress the development timeline while building a quality gate process to ensure every output met accuracy and inclusivity standards before publication.
Process
I started by mapping the development workflow and identifying which tasks AI could accelerate without sacrificing accuracy: first-draft SOP prose, knowledge check question generation, scenario variation, and alt-text writing. Tasks I kept fully human: needs analysis, SME validation, tone calibration, and final accuracy review.
I built a reusable prompt library for each task type. For SOP drafting: a structured prompt including role context, audience, output format, tone constraints, and a mandatory "flag any assumptions" instruction. Each prompt went through 3 iterations before becoming a team standard. I documented all prompts in a shared library for team reuse.
I designed a 12-point evaluation rubric covering: factual accuracy (SME-verified), gender/cultural representation in scenarios, reading level (target: Grade 8), absence of jargon assumptions, and alignment to company style guide. Every AI-generated artifact was scored before entering production.
I documented the full comparison: unedited AI output, rubric score, edits made, rationale for each edit, and final output. This transparency layer was critical for building stakeholder trust in AI-assisted content — and for training junior designers to use the same process.
Deliverables
Results & Takeaways
The AI-augmented workflow reduced first-draft time by approximately 58% across the SOP set. However, the evaluation and editing phase added back roughly 20% of that time — meaning the net gain was around 38%, concentrated in early-stage drafting. The most valuable output wasn't the time savings: it was the prompt library and evaluation rubric, which gave the whole team a repeatable, auditable process.
AI doesn't replace instructional design judgment — it accelerates the parts that don't require it. The designer's job shifts from content generation to content curation and quality assurance. That's a higher-leverage use of design expertise, not a lesser one.
Crisis Escalation Tabletop Exercise
A high-stakes facilitated simulation for an escalations team — scenario design, branching decision tree, Storyline interactive module, and structured debrief framework for a brand-safety incident.
Overview
A tech platform's escalation team handles YouTube's most sensitive brand and safety incidents — coordinated harassment campaigns, viral misinformation, and creator account crises — often after hours, with rotating on-call staff who may not have handled that specific incident type before.
A post-incident review revealed that response time variability was highest among staff with less than 12 months of experience — not because they lacked knowledge, but because they'd never practiced the decision-making under pressure that high-severity incidents require. Knowledge alone doesn't prepare someone for a 2am escalation call.
This project designed a tabletop exercise and Storyline simulation that put learners inside a realistic brand-safety incident and forced real decisions — with visible consequences.
Process
Working with escalation managers, I designed a fictional but realistic scenario: a high-profile creator's account is incorrectly flagged during a coordinated report attack, triggering a brand crisis at 11pm. The scenario had 4 decision points with realistic tradeoffs — not obvious right/wrong answers, but genuine judgment calls with downstream consequences.
I mapped all branches on paper before opening Storyline: 4 decision points, 2–3 choices each, with consequences that compounded across the scenario. I used Storyline variables to track whether learners followed proper escalation protocol — their final debrief reflected their specific path, not a generic summary.
For the live tabletop version, I designed a facilitator guide with: scenario setup script, facilitator decision cards (to introduce complications mid-exercise), group debrief questions organized by decision point, and a post-exercise action planning template. The guide was designed for a facilitator who hadn't written the scenario.
I built the interactive version in Storyline 360 with branching consequence feedback, variable-driven personalized debrief, and a "replay path" option letting learners revisit any branch. Published as SCORM 1.2 with an 80% completion threshold; tested in an LMS sandbox across desktop and mobile.
Deliverables
Results & Takeaways
The scenario's most effective design decision was making consequences compound across decision points. A poor choice at decision point 1 didn't just generate negative feedback — it changed the available options at decision points 2 and 3, reflecting how real incident management actually works.
High-stakes practice scenarios need to feel consequential — not punitive. The goal isn't to make learners fail; it's to give them a safe environment to experience the weight of a decision before it matters. That requires scenario design that mirrors real ambiguity, not artificial clarity.
Data-Driven Redesign: From 34% to 79% KB Helpfulness
A full analytics iteration story — KB effectiveness data revealed a structural failure, root cause analysis led to a targeted redesign, and post-launch metrics confirmed the impact.
Overview
A content operations team used an internal KB to handle policy edge cases — judgment calls that required consulting a reference article before making a decision. Quarterly analytics showed that the helpfulness rating on one high-traffic article cluster had dropped to 34%, with a bounce rate of 68% and an average time-on-page of 22 seconds — not enough to actually read the content.
Leadership assumed the content was wrong. My root cause analysis revealed a different problem: the content was accurate but structured for someone who already understood the policy, not someone actively trying to apply it under time pressure.
Process
I pulled helpfulness ratings, bounce rates, time-on-page, and search query data for the affected article cluster. I also reviewed 40 support tickets that referenced these articles to understand what questions users were actually trying to answer. The pattern: users were looking for a decision, not an explanation.
The articles were organized by policy section (legal/logical structure), but users approached them by scenario (operational structure). A user asking "Can I approve a borderline copyright claim?" had to read three separate articles to triangulate an answer. The information existed — the architecture failed to surface it at the right moment.
I restructured the article cluster around decision scenarios rather than policy sections. Each article started with a plain-language decision statement, followed by a 3-step decision tree, then supporting detail for edge cases. I also built a companion Rise microlearning module that walked through the same 3 scenarios with practice decisions — used as onboarding reinforcement.
I established a 90-day measurement window post-launch. Success criteria: helpfulness rating above 65%, bounce rate below 40%, and a 15% reduction in support tickets referencing these articles. I documented the methodology so stakeholders could replicate the evaluation cycle quarterly.
Deliverables
Results & Takeaways
Helpfulness rating reached 79% at 90 days (from 34%). Bounce rate dropped to 31%. Related support ticket volume fell 22%. The most important outcome was the measurement methodology itself — leadership now had a repeatable process for evaluating KB effectiveness quarterly, not just when something visibly broke.
The data told us what was broken. The root cause analysis told us why. Without both, the redesign would have fixed the wrong thing — better-written content in a structure that still failed users. Analytics without diagnosis is just confirmation bias with numbers.
Blended Onboarding Program for Data Center Operations Technicians
Full program architecture for a technical ops role — pre-work through 90-day milestone, designed for global scale and modular reuse across geographies and languages.
Overview
A data center operations team was onboarding 10–30 new technicians per month across three geographies. The existing onboarding consisted of a 2-day ILT session and a shared drive of PDFs — with no structured self-paced learning, no performance support, and no milestone assessment to confirm readiness before technicians worked independently on critical infrastructure.
My task was to design a full blended learning program from pre-arrival through 90 days on the job. The program had to be modular enough to localize for three regions, scalable enough to run without direct L&D involvement in each cohort, and rigorous enough to meet safety and compliance requirements for critical facility work.
Process
I worked with senior technicians and operations managers to map the 23 core tasks a data center technician performs in their first 90 days — by frequency, criticality, and consequence of error. This became the backbone of the learning architecture: high-frequency + high-consequence tasks got the deepest treatment; low-frequency tasks got job aids.
I designed a 5-phase blended program: (1) Pre-arrival self-paced modules covering safety fundamentals and site orientation; (2) ILT days 1–2 focused on hands-on lab and team integration; (3) Self-paced eLearning weeks 1–4 covering technical procedures; (4) On-the-job performance support (job aid library for the 8 highest-consequence tasks); (5) 30/60/90-day milestone assessments tied to readiness-to-work-independently criteria.
I designed the 2-day ILT with a facilitator guide that any trained technician could run — not just L&D staff. The guide included timing, materials lists, discussion prompts, lab activity instructions, and debrief questions. The ILT was designed to be identical across all three geographies with only localized examples.
I built the eLearning content as independent modules (not a single course) so regional teams could swap individual modules when procedures changed without rebuilding the whole program. Each module followed a consistent template: objective statement, procedure walkthrough, two practice scenarios, knowledge check, and job aid link. This modularity also enabled localization without redesign.
Deliverables
Results & Takeaways
The modular architecture was the single most important design decision. When a safety procedure changed 4 months after launch, the team updated one module and redeployed it — without touching the ILT guide, job aids, or any other component. That's what scalable content architecture actually looks like in practice.
Blended learning programs fail when the modalities don't connect. The job aids had to reference the eLearning; the ILT had to reinforce the pre-work; the 30-day assessment had to reflect what the eLearning actually taught. Designing the connections between modalities is as important as designing each modality itself.
[Your Name]
I'm a senior learning experience designer who builds learning systems that solve real operational problems — not just courses that check compliance boxes. My work spans knowledge base architecture, AI-augmented content development, high-stakes simulations, and blended onboarding programs for global tech teams.
Before moving into instructional design, I [your background]. That experience gave me a ground-level understanding of how people actually perform under pressure — which shapes every design decision I make.
I completed the LXD & ID Bootcamp at HeeRise Academy and hold [your certifications]. I'm focused on roles where learning design is treated as a strategic function, not a support service.
I'd love to connect with you.
Interested in learning experience design roles at tech companies and global operations teams.