[Observability AI Assistant] Better title for Conversations generated by the Rule Connector

[emma-raffenne]

Context

An Observability AI Assistant connector is available to be set as action for Observability rules. When an alert is triggered, a conversation with the AI assistant will be created sending the initial prompt set by the user in the rule action. The conversation is then stored and can be retrieved from the AI Assistant interface.

Problem

The evaluation conducted on the AI Assistant connector revealed that conversations created from the connector are saved always with the same title, i.e. "Understanding and Responding to Alert Triggers". Since more than one conversation can have been created in the same way, and there is no indication of the creation date or of the creator of the conversation in the interface, it is impossible to recognise them.

Suggested improvement

For conversations created from an alert rule connector, the following pattern could be used: {rule title} - {timestamp}

[emma-raffenne]

cc @boriskirov

[viduni94]

Won't fix at the moment due to this reason:
#198791 (comment)

[emma-raffenne]

Reopening as titles may still appear duplicated. Pending UX design input to be addressed.

[boriskirov]

Can we either use:
-а randomly generated key with a dash to differentiate if we don't have any metadata to use: {name}-{random-key}

• or request from the LLM to always generate a different name, if there is already existing one

@viduni94

[viduni94]

Can we either use: -а randomly generated key with a dash to differentiate if we don't have any metadata to use: {name}-{random-key}

• or request from the LLM to always generate a different name, if there is already existing one

@viduni94

Hi @boriskirov

I don't think I understood what you meant by if we don't have any metadata to use. (We would always have a timestamp, when the conversation is initiated)

For the 2 options you mentioned:

• {name}-{random-key} - We could use this. However, the problem we were discussing about the timestamp was similar to this. If we have a long title, it wouldn't translate to the unified UX (when we implement it), because long titles are truncated in the unified UX design: More context here
• request from the LLM to always generate a different name - I tried this, however the different names generated by the LLM are not very different and doesn't help much to uniquely identify the conversation (More context here)