Executive summary

Project example: SETREX app · Source material: approx. 27 minutes of screencast · Status: November 2026. Case study – documented work in progress, not a statement about completed live operation.

A SETREX screencast of around 27 minutes serves as the starting point for a reusable knowledge and training system. Instead of merely publishing the recording, its content is transcribed, structured into six process units and prepared for different situations of use: learning, looking things up, revising and, in future, dialogue-based knowledge queries.

The work so far has produced the thematic structure, six learning modules, presentations with screenshots, module handouts and dialogue-style podcast scripts. Multiple-choice tests are being designed; a wiki and an AI chatbot are planned. A final sign-off by subject-matter experts has not yet been documented.

Initial situation and challenge

A classic screencast shows clearly how a system is operated, but it remains linear. Employees have to search for the relevant passages; the content is hard to look up, learning success is not checked, and changes to the process are difficult to update module by module.

  • Knowledge sits in image and speech, not in a structured knowledge base.
  • Different target groups need different formats: short learning paths, reference texts or answers to specific questions.
  • The original voice is not strictly necessary for neutral training that can be reused in the long term.
  • Factual statements must be checked against the actual SETREX process before release.

Target picture: one piece of content, several formats that can be validated

The source is not only transcribed but transferred into a traceable content structure. This structure is the common basis for all subsequent formats.

Screencast → transcription → topic and process structure → comparison with the visual material → modular learning content → presentations, handouts, podcast scripts → in future tests, wiki and RAG chatbot.

The formats have different jobs: learning modules guide users through the process; handouts support looking things up; tests check understanding; a wiki offers reference information that can be versioned; a RAG chatbot is intended to answer questions on the basis of approved content and name its sources.

Implementation using the SETREX example

  • Module 1, Getting started and sign-in: registration, company activation, selecting the company, QR onboarding, bookings and status.
  • Module 2, Shooting day overview and check-in: settings, synchronisation, shooting day details, booking tickets and manual check-in.
  • Module 3, Check-out and follow-up: manual check-out, Setpass QR, time corrections, breaks and additional costs.
  • Module 4, Pay settings and contract conclusion: remuneration, costs, supplements, roles, notes, cancellations and electronic signature.
  • Module 5, Special cases and reporting: break settings, batch signature, street casting, Excel cost report and ZPass scans.
  • Module 6, Terminal mode and QR codes: kiosk mode, QR configuration, self-registration and check-in without prior booking.

The modules follow the operating procedures covered in the source material. Their final factual accuracy and currency need to be confirmed separately.

Media production and instructional design

Presentations and handouts: suitable screen sections from the screencast are assigned to the process steps. The presentations show the user interface and the context of each action; handouts summarise procedures in a reference form that is quick to use.

Spoken content and podcasts: the original spoken text is revised linguistically and can be re-recorded with a neutral synthetic voice. In addition, dialogue-style podcast scripts with a female host and a subject-matter expert were developed to explain the process logic and typical follow-up questions clearly.

Knowledge testing and dialogue-based access: multiple-choice questions are to be derived from the learning objectives approved by subject-matter experts, with unambiguous correct answers and well-founded distractors. A wiki and a RAG chatbot are intended as further output channels, but should not be understood as already implemented.

Data protection and neutralising the speaker’s voice

Replacing the original voice can make the speaker less recognisable and create a consistent training voice. On its own, however, it does not ensure GDPR compliance. Screen recordings, names, contact details, booking information or other personal data can also be relevant.

  • Check visible personal data in screenshots and videos and anonymise it where necessary.
  • Assess the legal basis, purpose, permissions, retention and the AI service providers used separately.
  • Document approvals and versions of the media used.

Validation and quality assurance

A reliable knowledge system needs a multi-stage review. The separation between fidelity to the source and factual correctness is especially important: reproducing the screencast correctly does not yet prove that the original process is current or factually right.

  • Source check: is the statement covered by the transcript and the visible operating procedure?
  • Expert review: does it match the currently valid SETREX function and the company’s operating rules?
  • Format consistency: are learning module, handout, audio, test and knowledge entry free of contradictions?
  • Didactics and usability: are procedure, terms and test questions understandable and action-oriented?
  • Approval and versioning: who approved which version when, and how are changes maintained?

Result and outlook

Project status based on the documented work:

  • Transcription and topic structuring: developed
  • Six learning modules: developed
  • Presentations with screenshots: created
  • Module handouts: created
  • Conversational podcast scripts: created
  • Multiple-choice tests: in design
  • Wiki and AI chatbot: planned
  • Expert quality assurance and sign-off: not yet documented

The next sensible step is a review and sign-off round with the people responsible for the SETREX process. On that basis, multiple-choice tests, wiki entries and a citable knowledge base for the chatbot can be completed consistently. After that, usage data and learning results can be compared before and after the introduction.

Conclusion

The real added value lies not in automatically generating as many media as possible, but in turning existing process knowledge into a structured, verifiable and updatable knowledge base. The SETREX case shows the path from a one-off screencast to modular employee training and, in future, to knowledge queries as needed. Quantitative effects on time, cost or learning success have not yet been measured reliably and are therefore not claimed.