Stop before the expensive part
Models write confident, wrong training content. The pipeline stops at the storyboard so a subject expert can correct it first.
One brief goes in. After one human review, a finished, narrated training module comes out.

1 minute 1 second, with sound and captions.
On screen: A film leader counts 3, 2, 1 in the gate of a strip of film. The words on it: Approved for anyone with a deadline. A Sheldon Smith production.
[Upbeat jazz-funk music plays] [A tone sounds on the film leader]
On screen: The opening screen of a finished course, Closing the Till the Blue Palm Way, sits in the gate. It is labeled as a real build of a fictional retailer’s till-closing course, screen 1 of 18, and the words of its narration light as they are spoken.
Course narration: Welcome. Every shift ends with a moment that matters more than most people realize: the close.
On screen: The film moves on a frame at a time through more of the course: flip cards, and one flips; a hotspot; a knowledge check; a branching scenario, and a choice is made.
[Film clicks through the gate]
On screen: Giant type fills the frame: 18 screens, one brief.
On screen: The title, E-Learning Pipeline, runs in on a strip of film, a letter to a frame. Under it, a card shows the first sentence of the brief, and three lines land: One brief goes in. One human review. One course comes out.
[The full band comes in] Give me one brief. After one human review, I hand back a finished, narrated course.
On screen: The plan. A storyboard of 18 blank, numbered frames lays out on three strips while a clock, labeled an estimate, counts down from ten minutes. Stop: nothing else gets built, and the clock holds. A hand brings down a stamp, Approved by a person, and the clock runs again.
I write the storyboard first. Then I stop. Nothing else gets built until a person approves it.
On screen: The crew. The 18 frames fill with their screens, then five panels run in, one over another, each with a real screen and a name: the screens, the knowledge checks, the branching scenarios, the illustrations, the narration. Beside the illustrations stands the course’s shop associate, stepped out of her picture as a paper cut-out. The clock reaches zero.
Then I build the rest: the screens, the knowledge checks, the branching scenarios, the illustrations and the narration.
On screen: Red. The word for Arabic in Arabic script, the words Right to left, and the course player’s own controls as the renderer draws them in Arabic. A film can rolls in: SCORM 1.2, ready for an LMS. A line of Arabic narration is written out, with its meaning: Module complete.
I do it in Arabic too: right to left, narrated in Arabic. Then I package it as SCORM. Arabic narration: (In Arabic) Module complete.
On screen: By Sheldon’s estimate: Ten hours is a job best done in ten minutes. Each line is inked as it is said, and the paper associate walks on carrying a bag.
By my maker’s estimate, ten hours of work by hand takes me about ten minutes.
On screen: Two panels close like doors. On the left, the poster for The E-Learning Pipeline. On the right: The E-Learning Pipeline, now showing. A Sheldon Smith production, story by one brief, screenplay by Claude, narrated in English and Arabic, packaged as SCORM, running time ten minutes. More than 3,000 automated tests.
The E-Learning Pipeline. Now showing. [The band climbs] [The band plays a run of short hits] [One last hit, and the chord rings out]
This film did not load. Its transcript is on the project’s page. Read the transcript
The captions did not load. The transcript is on the project’s page. Read the transcript
The pipeline turns a brief into an interactive e-learning module. It writes a storyboard first, and nothing else is built until a person approves it. Then it produces the screens, knowledge checks, branching scenarios, illustrations and narration, and packages the result as SCORM 1.2 for an LMS. Every line stays editable in the browser until export.
By my estimate, a module that takes about ten hours to produce by hand takes about ten minutes.
Arabic is supported end to end, with right-to-left layouts and Arabic narration.

Models write confident, wrong training content. The pipeline stops at the storyboard so a subject expert can correct it first.
Every screen is checked against a strict schema, then repaired or rejected, before it can reach a learner.
The repo carries more than 3,000 automated tests.