Goal
Reimagine how we guide the user to make the process more interactive and prompt users to think about the end use case.

Problem statement

Making Knowledge Contribution Easier for SMEs

Right now, we ask subject matter experts to contribute knowledge to the AI by creating QnA pairs from selected contexts of chunked documents. The current experience, however, doesn’t give them a lot of guidance on what to do, how to do it well, or why it matters.

According to feedback obtained from the field specialists, creating qna.yml manually is hard and error-prone. Without the right prompts, examples, or context around what makes a strong QnA pair and how it ties back to their use case or end-user, users get stuck or second-guess whether their contributions are useful and accurate.

We want to make this process feel intuitive and supportive by:

• Breaking the task down into clear, manageable steps
• Providing helpful examples and tips along the way
• Reminding them of the real-world use case their work is supporting
• Making it easy to adjust, edit, or select from QnA suggestions
• Helping them feel confident that what they’re doing is correct and impactful

Background

• As an example: Have a user answer questions around ‘what are questions the trained model will absolutely need to answer?’ (i.e. ‘what are questions your users will absolutely ask’) - make the process more ‘interactive’ - prompt users to think about the end use case
• How can we provide a domain-specific experience for use cases like, what if it is a chatbot or app for one of these: https://writer.com/customers/
• Prompting improvements (where can we start minimally and encourage this in the UI) while we wait for some of the experiments to prove out:
  • Use Case
  • How you should be thinking about writing a QnA.YAML
  • How can we provide the guidance in a different way
  • Kinds of questions and context to select

Considerations

• We should prompt this sentiment from the beginning but we would consider breaking this up to focus on document creation (content) and seed example prompting (QnA format, how the model responds to the content)

Take aways from Demo on April 3 from duffy@redhat.com:

• The tooling should let the user undertand very clearly at each step: (1) what content gets into the model; (2) what format is the model going to process the content
• If you want a specific type of content to get it into the model, the user does not need to get it into the chunks for QnA, as long as it is in the full SDG data set
• Select context based on the types of QnA you want answered
• It is more about the format of the chunk or the method you want to interpret it

Resources

• Slack thread about prompting the user: https://redhat-internal.slack.com/archives/C08AQC1C29H/p1743181803009269
• Field feedback document: https://docs.google.com/document/d/1mWxbWv14Eq1oiBB2yuKQ9JU48msmG-pH4xQwEq12JHc/edit?tab=t.0 
• Miro board for documenting flow and lo-fi sketches: https://miro.com/app/board/uXjVIGteSO0=/ 
• Demo deck 4/17: https://docs.google.com/presentation/d/14HZwtGAtoGYyhkg1RdwjcihGDUMjJD44J7LgugbtOAs/edit?slide=id.g34c65c13846_0_2676#slide=id.g34c65c13846_0_2676 
• Demo deck 4/10: https://docs.google.com/presentation/d/1EXCsodockerYG0P5QV0jAOJpgVB4y_hEdu6OPvDbsjQ/edit?slide=id.g34c65c13846_0_2676#slide=id.g34c65c13846_0_2676 
• Figma, mid-fi wires: https://www.figma.com/design/TZO1xQUVQLrZzEIItDbTmb/Stage-04?node-id=143-31749&p=f&m=dev 

Deliverables
[Fill in]

Next Steps

• rhn-support-amaredia set up a call with SDG folks, UX and PM to align on a workflow