Hi Community,
I've been exploring Rovo recently, and one question keeps coming to mind.
Let's say I ask Rovo something that could be answered in a few different ways:
- Using information from our Jira or Confluence content.
- Using the LLM's general knowledge.
- Or by combining both.
How does Rovo decide which source to rely on?
For example, does it always search the indexed content first and then use the LLM to generate the response? Or can it sometimes answer directly from the model without looking at organizational data?
I'm also curious about what happens if the information in Confluence or Jira contradicts what the LLM already "knows." Which one takes priority in that case?
And finally, is there any sort of confidence score or retrieval threshold that determines whether Rovo uses retrieved content or falls back to the model's own knowledge?
I'm asking because understanding this would make it much easier to write better prompts and, more importantly, know how much to trust the responses in real-world enterprise use cases.
Would love to hear from anyone who's looked into this—or if someone from the Atlassian team can share how this works at a high level.
Thanks! 🚀
