The challenge

The subject matter experts brought a thorough slide deck: definitions and common myths, real cases from other organizations, consequences for patients, employees, and the organization, controlled substance access points, warning signs, and UCSF's reporting pathways. Everything a learner needed was there, but it was organized as information. The job was to turn it into the moments where someone has to notice something and act.

What I did

• Reorganized the content around two skills: recognizing warning signs and reporting a concern.
• Created scenario videos with Google Flow, an AI video tool. In one, a nurse is asked to witness a medication waste she didn't see. In another, a pharmacist notices signs that a colleague is impaired.
• Followed each scenario with an interactive knowledge check, so learners make the call before the course explains it.
• Added AI audio narration throughout.
• Closed with an 11-question knowledge check (80% to pass) and a policy attestation, then set up the course and its audiences in the UC Learning Center.

Why it works

AI video made realistic clinical scenarios possible without a film crew or patient footage. Learners practice the uncomfortable part, deciding whether what they saw is worth reporting, in a setting where getting it wrong is safe.

The challenge

Data Stewards decide how research data can be shared. They have to apply UCSF Policy 5.01.06 correctly while following the data request process step by step. The source material was an SME slide deck and script that covered the policy in depth. Analysts needed to leave able to handle a request, not recite the policy.

What I did

• Worked with the SMEs to separate what analysts need to know from what they need to do.
• Interviewed analysts about their actual workflows.
• Built scenarios from those interviews, so learners apply the policy to requests like the ones that land in their queue.
• Developed the module in Storyline with AI audio narration and published it as SCORM in the UC Learning Center.

Why it works

The need-to-know and need-to-do split kept the module short enough to finish and focused on decisions. Grounding scenarios in analyst interviews meant the practice matched the job. This was my first project at UCSF.

Try it

The bot is live and protected by an access code to keep costs predictable. Email me for the code and I'll send a link.

The problem

Recruiters often have a quick question that doesn't justify a call. Which learning platforms have I administered? Have I built clinical training? I wanted those answered at any hour, accurately, and without the tool overselling me.

How it works

1. AccessA visitor enters the access code. The server applies rate limits.
2. QuestionThe Express server sends the question to Claude with a curated fact sheet.
3. AnswerClaude answers only from the fact sheet, which is drawn from my career records.
4. Hand-offIf the answer isn't there, it says “I don't know, please ask Jeff” and offers my email.

Guardrails I designed in

• It never invents metrics, and it identifies itself as an AI assistant.
• Private details stay out of the fact sheet entirely, so the bot can't reveal them.
• The API key lives only in the host's environment settings, and the app runs in its own API workspace with a spend limit.
• I test it against a set of questions, including trick questions about work I shouldn't get credit for, after every change to the fact sheet.

Why it matters for learning work

The same pattern of a small server, a curated knowledge source, and strict answer rules is how I'd add AI to eLearning: scoring open-ended responses, branching scenarios on what learners actually write, or answering questions inside a course without drifting from approved content.

The challenge

The content is policy heavy and serves three main, overlapping audiences: adult, pediatric, and the community hospitals. It covers five locations: Parnassus, Mount Zion, Mission Bay, BCH Oakland, and Stanyan/Hyde. A single linear course would make everyone sit through rules that don't apply to them.

What I did

• Co-built the location mechanism so each learner gets the content for their site.
• Built the course's topic sections on Earthquakes, Safe Patient Handling, Aerosol Transmitted Diseases, Bloodborne Pathogens, Occupational Exposures, COVID-19 Resources, and Hazardous Materials and Spills.
• Built each slide from a shared layout system (title, side image, two and three columns, side panel, list) defined in the course style guide, so three designers' work reads as one course.
• Worked stakeholder feedback through a tracked feedback log and a crosswalk of review transcripts, resolving items and recording each decision so reviewers could see what changed.
• Built a branded HTML version of the training site with carousel and mini-manual interactions.

The challenge

Child Life Services had taught this material as a live mini-workshop series. They needed every radiology staff member trained, including new hires who missed the workshops. The strongest material was the Child Life specialists' own voices, but their recordings had background noise and uneven levels.

What I did

• Turned the four workshops into four chapters: the Child Life Specialist's role, the One Voice approach, language considerations, and SWAP, the Scan Without Anesthesia Program (with Feed and Swaddle for infants).
• Wrote objectives around what staff do in the room: use developmentally appropriate language, keep one voice during a procedure, and support families through a scan without sedation.
• Embedded video demonstrations and added a knowledge check to each chapter, importing the question banks into Storyline from spreadsheets.
• Used ElevenLabs to clean up the specialists' recordings, removing background noise and evening out levels, so their voices could stay in the course.
• Set up the LMS assignment audience and the assignment, reminder, and completion notifications.

Why it works

The One Voice approach is simple to state and hard to do: a quiet room, one person speaking, parents involved, and a prepared child. Hearing the specialists explain it in their own words, then seeing it demonstrated, makes it concrete for staff who have never watched a Child Life Specialist work.

The challenge

UCSF's HIPAA training was a long annual slide deck. People click through it once a year and forget most of it. The Health Privacy Office wanted training that was more engaging and easier to read, and that separated what clinicians need from what researchers need.

What I did

• Drafted a proposal to release the training as four quarterly microlearning modules instead of one annual course, grounded in research on spaced repetition and the forgetting curve.
• Prototyped a game-style module in React: six modules from Fundamentals to California law, a hospital map of locations to explore, and points for completed scenarios.
• Mocked up scenario-based knowledge checks. In one, a coworker asks you to look up a friend's lab results, and you explain what you would do and why.
• Started the first Storyline scenario, using AI-generated images and video for the setting.
• After the Privacy Office's design review, planned the branching that sends clinicians and researchers to different content, and how the LMS could assign the research path automatically.

The series

1. Outpatient E&M Services35 minNew vs. established patients, coding by medical decision making or time, telephone, video, and e-visits.
2. Inpatient and Observation E&M Services36 minSame-day admission and discharge, transfers between sites of care, and critical care.
3. Surgical Coding for Minor Clinic and Bedside Procedures39 minMinor vs. major surgery, finding the right procedure code, modifiers 24, 25, and 59, and Medicare-specific modifiers.

The challenge

The compliance educator had three detailed slide decks full of coding rules, CPT code ranges, and modifier guidance. Providers new to MarinHealth needed to learn to code their own work accurately, and they would be audited on it. The courses had to be short enough for busy clinicians and consistent across three builders.

What I did

• Built the Surgical Coding for Minor Clinic and Bedside Procedures course in Storyline from the educator's slides and narration.
• Developed the shared Storyline template for the series, including accessibility buttons and player instructions. It later became the team's accessibility template.
• Turned the educator's knowledge check documents into scored checks with a rationale for each correct answer.
• Closed each course with what to expect from the Self-Coding Audit Program and where to get help.

The challenge

Every Storyline course had to meet WCAG 2.1 AA, and each designer was solving the same problems slide by slide: player instructions, keyboard shortcuts, playback for screen reader users, tab order, and quiz settings. The fixes were inconsistent, and a single missed trigger could lock a keyboard user out of a quiz.

How it's organized

1. Scene 1WelcomeTitle screen. Always manual navigation.
2. Scene 2 · RequiredAccessibilityTraining Information, Accessibility Information, Playback Mode, and Autoplay Instructions. Never deleted.
3. Scene 3ContentImported PowerPoint slides go between two anchor slides, with triggers set by slide type.
4. Scene 4QuizIntro, questions, results, and end screen. 80% to pass by default.

Learners choose how slides play

An Autoplay button in the corner switches modes at any time. Training Information (Ctrl+Alt+I) and Accessibility Information (Ctrl+Alt+A) are reachable from any slide, with a Return button back to where the learner was.

Built-in guidance for designers

• “Delete before publishing” instruction slides open each scene and walk the designer through it, covering three course types: no audio, voiceover, and video.
• Five rules that are never broken: no auto-advance on quiz or interaction slides, no manual-play trigger on silent slides, never delete the accessibility scene, always set tab order, and always review captions.
• A quiz checklist: shuffle off when “All of the above” is used, no letter labels, a rationale for every correct answer, keyboard-only testing, and pass or fail stated in text, not color alone.
• AI where it saves time: ChatGPT Enterprise formats question banks for bulk import, the Storyline AI Assistant drafts alt text for review, and Adobe Podcast cleans up client audio.
• A demo package with a blank build, a completed build, source slides, and a question bank, so colleagues can practice the workflow.

Where it came from

The template started as the shared template for the MarinHealth self-coding series. Once the accessibility pieces were working there, I generalized it for every course type the team builds.

The challenge

Early Start ends when a child turns three. Families and the professionals who support them then have to work out whether the child qualifies for regional center services under the Lanterman Act, what other community resources exist, and who is responsible for each step. The rules come from state code and change over time.

The modules

1. Module 8: Lanterman Eligibility and PlanningEligibility requirements, assessment procedures, and service plan development for regional center services.
2. Module 9: Community ResourcesSupports available to families beyond the regional center.
3. Module 10: Roles and ResponsibilitiesWho does what during the transition out of Early Start.

What I did

• Designed and produced the narrated modules in Camtasia from subject matter experts' materials, breaking the transition process into the decisions families and providers actually face.
• Cited the governing code on screen, such as Welfare and Institutions Code 4512(a)(2), so learners can find the source.
• Delivered the courses through the Totara/Moodle LMS I administered, with evaluation collection automated through IntelliBoard.

The challenge

A small training team served a whole university. It needed evidence about which courses to build, which to deliver live, and which needs were already met by existing resources, and it needed that evidence broken out by division.

Response

• 208 of 634staff completed the survey (33%)
• 51 of 108managers and leads completed the survey (47%)

How it worked

• With my co-author, designed separate surveys for staff and for managers and worked with Institutional Research, which administered them under an IRB protocol, so no identifying information reached the training team.
• Asked staff to rate the interpersonal and technical skills their jobs require, and their interest in 24 potential courses, on a five-point scale.
• Ranked responses by weighted average and broke them out by division using federal job categories.
• Met with division leadership teams to review the data and hear their priorities directly.

What we recommended

• Meet the top requests, Microsoft Office and Google Apps, by promoting the self-paced courses already in the SkillPort LMS, which saves trainer time for topics unique to the university.
• Pair high-ranked technical topics like succession planning and documenting work processes with change management in a single course series.
• Build soft-skills courses where staff interest and leadership priorities agreed, such as cross-cultural communication and negotiation.
• Create a university-wide wiki for documenting processes and job duties.