Back
Get in touch
EquipaWorks Case study · Learning · AI automation

A scalable system for automated learning content

One topic in, one complete microlearning video course out, in about 2 hours, with two human-in-the-loop checkpoints.

~2 hper course
13courses
78videos

Self-initiated product, 2026. Built to test AI safety training for small clinics.

The published EquipaWorks course page.
A frame from an EquipaWorks lesson video.
Role
Sole builder: research, learning design, automation, video editing, brand design, QA and pilot
Deliverables
  • 2 Make scenarios: write the course, then make the videos
  • Airtable control panel and QA log
  • Custom Synthesia templates (intro and lesson)
  • 13 courses: 78 videos, 13 quizzes, 13 course guides, 13 job aids
  • Branded learning site
  • Case study (PDF)
Built with
MakeAirtableOpenAISynthesiaClaude Design
01Problem

AI was coming into clinics.

No one was training care teams to use it safely.

What I saw

At a Fortune 500 company, I did field research in clinics across the US and saw care teams stretched between technology, workload and regulations like HIPAA.

The bet

Small clinics would need practical AI safety training.

To test it, I needed real courses at scale, without a production budget.

02The course

One topic in. A complete course out.

Microlearning for the moment of need: short video lessons, practice in every lesson, and a job aid to use on the job.

The real quiz from this course, generated from the lesson scripts.

Loading the quiz…

The quiz file (equipaworks-quiz.json) hasn’t been added yet.
{{ qProgress }}

{{ qText }}

{{ qFeedback }}
Your score {{ qScore }}
Loading…
{{ pdfName }} hasn’t been added yet.
Course · 9 parts

Safely Summarizing Medical History with ChatGPT

Every lesson follows the same 10 parts
action and practice
  1. 01Hook
  2. 02Welcome
  3. 03What you’ll learn
  4. 04Why it matters
  5. 05Core idea
  6. 06What to do
  7. 07Real-world example
  8. 08Your turn
  9. 09Key takeaway
  10. 10Outro

Why microlearning: a review of 17 studies found it improved knowledge, confidence and retention in health professions education (De Gagne et al., 2019). A meta-analysis of 217 studies found practice testing beats re-reading (Adesope et al., 2017).

03How it works

Airtable is the control panel. Make does the work.

Topic in
Step 1Write the courseMake + OpenAI, 8 AI calls
Human-in-the-loopScript review
Step 2Make the videosMake + custom Synthesia templates I built
Human-in-the-loopVideo QA
Publishto the learning platform

A person also chooses every topic before the system starts.

04Impact

Faster to make. Faster to change.

Traditional production This system
Make one course
Traditional9–15 hours to script, produce and edit
This system~2 hours, with two human checkpoints
Cost per course
Traditional$310–$485+
This system~$135–$155
Update a lesson
TraditionalReshoot, re-edit, re-upload: days
This systemEdit the script, re-render, replace: minutes. Same link, every embed updates.
Consistency
TraditionalVaries by writer and editor
This systemSame 10-part structure and the same checks every time

Traditional: Chapman Alliance benchmark, 49–79 hours per finished hour of basic e-learning, at $33.31/hour (BLS 2025 median, training and development specialists). This system includes a full year of tools on annual plans.

The result

50–70% lower cost, and a library that keeps up when the rules change.

Rules keep moving

California AB 3030 already requires disclaimers on AI-generated patient messages, and a major HIPAA Security Rule update is on the way.

05Proof

What I checked, and what I missed.

Video QA17/ 78

Videos flagged in QA. All were avatar issues, and all were re-rendered before release.

Structure64/ 65

Lesson scripts had all 10 parts.

Honest miss0/ 65

Lessons hit the word target (median 310 words vs 170–220), even after several prompt rewrites.

What QA flaggedIssues across 17 videos
  • Eyes10
  • Glitch5
  • Body movements2
  • Voice or pronunciation2
  • Background noise1
  • Looking down1
Takeaway

The bottleneck was the avatars, not the writing.

Every flagged video was an avatar rendering issue. Scripts held their structure in 64 of 65 lessons.

Flagged videos caused by avatars17 of 17
Eyes, glitches, movement and voice
Scripts with all 10 parts64 of 65
The writing held up
06Pilot

The need was real. The timing wasn’t.

01
Saw the gap in the field
02
Built the system
03
Made 13 courses
04
Piloted with 6 clinics
05Too early
Ended the project
Next
What could change the timing
04Why clinics weren’t ready

They were still learning to trust the AI tools already rolling out. Training felt like more work, not relief.

05Why it was still a win

Courses were cheap to make, so the answer took weeks, not a production budget.

NextWhat could change the timing

The HIPAA Security Rule update, now expected to be final in July 2027, would require documented security training at least every 12 months and an inventory of AI tools that touch patient data.
  1. 01
    Saw the gap in the field
  2. 02
    Built the system
  3. 03
    Made 13 courses
  4. 04
    Piloted with 6 clinics
    Why clinics weren’t ready

    They were still learning to trust the AI tools already rolling out. Training felt like more work, not relief.

  5. 05Too early
    Ended the project
    Why it was still a win

    Courses were cheap to make, so the answer took weeks, not a production budget.

  6. Next
    What could change the timing

    The HIPAA Security Rule update, now expected to be final in July 2027, would require documented security training at least every 12 months and an inventory of AI tools that touch patient data.
What I learned
  1. 01Timing is part of product-market fit.
  2. 02Build the cheapest real test.
  3. 03Prompts can’t enforce limits. Word counts belong in a code check.
Sources

1. De Gagne et al., JMIR Medical Education (2019)2. Adesope, Trevisan & Sundararajan, Review of Educational Research (2017)3. Chapman Alliance e-learning benchmark (2010), via Christy Tucker4. BLS wage data (2025), training and development specialists5. Synthesia Help Center, video versioning6. Fierce Healthcare, HIPAA Security Rule timeline (2026)7. Faegre Drinker, proposed HIPAA Security Rule updates (2025)8. California AB 3030 summary, ArentFox Schiff9. Tool pricing: Synthesia, Make, Airtable and OpenAI plan pages (2026)

{{ darkAmbient }}
Go Deeper

See how it was built

Built to test AI safety training for clinics, and ready for any team that trains people in regulated environments.

{{ lbAlt }}
{{ lbFile }}
{{ lbTitle }}{{ lbCount }}