Trained Models nav updates

[mdefazio]

Updating this issue to reflect our current thinking and new navigation approach

Trained Models Nav Updates

Trained Models are a shared concern across solutions and need a permanent spot that makes sense for admins to manage. This work relates to the Inference Endpoint work.

We would like to move the current ML > Trained Models view into Stack management under a new section.
Transition to removing the current Alerts and Insights > Machine Learning page and move this workflow into Saved Objects

• This let's solutions avoid needing a special place (unless desired) within their navigation for Trained Model management
• Will also allow us to remove the duplication of Trained Models views in the app today (Alerts and Insights > Machine Learning)

Phase 1 tasks

Preview Give feedback

1. Move Trained Models into new section in Stack Management --> 'AI Ops'

Phase 2 tasks

Preview Give feedback

1. Make Machine Learning jobs available in Saved Object management #183660
   Team:ML Team:Platform-Design +2
2. Make Machine Learning trained models available in Saved Object management (Discovery task)
3. Research how job and model synchronization would work within Saved Objects management
4. Update Saved Object search and filtering to account for ML Jobs (and maybe Trained Models)

Phase 3 tasks

Preview Give feedback

1. Remove Alerts and Insights > Machine Learning
2. Update Saved Objects management to account for any additional needs for ML jobs (and models)
3. If needed, add space management to Trained Models view

---

Figma link

---

Stateful

Trained Model page in new Stack Management section

Classic nav

Solution nav

Serverless

Trained Models are available in Management > Trained Models

[mdefazio]

@ryankeairns @yanwalton I wanted to get this on your radar as far as nav changes go and what this may mean for managing saved objects across spaces (since this is the primary need for the current stack management > ML > trained models page).

The above designs are meant as proposals—I know there's work happening on navigation and this should of course fall inline with that.

And to reiterate, we will be creating a specific Inference Endpoints page to manage those endpoints created in Search with semantic_text mappings. This will live within the Search navigation, but link to Trained Models. (And ideally vice versa). But the effort to remove the confusion around multiple Trained Models pages is still relevant IMO.

[ryankeairns]

Just happened upon this, clearly I missed the notification 😬

For the serverless piece, do other solutions use Trained models? If so, that may make the case for placing it under Project settings stronger.

[mdefazio]

For the serverless piece, do other solutions use Trained models?

Yes. If i'm not mistaken, anyone needing to use ELSER for O11y assistant or Security AI would also potentially need to review that page to check the status of the model.

[mdefazio]

cc/ @arisonl @jgowdyelastic @peteharverson
I've tried summarizing our slack thread this morning into the issue description here.

[elasticmachine]

Pinging @elastic/ml-ui (:ml)

[arisonl]

Nit but not to forget:

Move Trained Models into new section in Stack Management --> 'AI Ops'
This should not be called AIOps, AIOps is a different thing. Search model MLOps maybe or something.

More comments to follow.

[mdefazio]

First phase scheduled for 8.16, but labeled as 'could'

Will need to account for Data Frame Analytics models separately and remain in the ML app (not Stack Management).

Rather than naming it 'AI Ops', currently planning on 'ML Ops' as the name for the section.

[mdefazio]

With this PR… does this get us close to removing the Stack Management > Alerts and Insights > ML > Trained Models page?

Basically jumping us to the Phase 3 tasks from the issue description?

[peteharverson]

With https://github.com/elastic/kibana/pull/201426… does this get us close to removing the Stack Management > Alerts and Insights > ML > Trained Models page?

@mdefazio no not really, this PR was to reduce the need to manually run the synchronization action when creating an inference endpoint. A separate piece of work is now in progress to move the Trained models list page, plus the ML job management lists, into Stack Management. I'll send you a link to the designs.