Hi,
I have an interesting use case that might be a pattern for non-LLM models. The concept I named as "Exhaustible overlapping topics", to name it in some way...
We have one Chatbot to support our developers on providing pointers to documentation, support requests, best practices, and a couple of more specific actions to raise tickets and incidents on our ticketing system.
The current way Power VA works is the following:
- You create different topics with triggering phrases
- Power VA tries to extract the best topic
- In case of different possible choices, it provides some suggestions.
Example:
What I have at the end of these topics is to allow rephrasing the question or raise a ticket, but in fact, any of these suggestions could help with the resolution. The "Exhaustible topics" would allow the user to select one of these, follow the steps and, instead of ending the topic as I referred before, come back to these topic suggestions, without (greyed out?) the already selected ones. That way, he can keep checking additional things without merging all these cases into a huge topic (which can be too broad if the request is more specific and PVA can find an exact match).
Once the user exhausts all choices, they can finish on the default ending (rephrase/ticket).
Is there any way to simulate this behaviour in any way? Otherwise, do you think this concept is valid for implementation in PVA?
Cheers!