Power Platform Community Forum Thread Details

I am creating a chatbot with Copilot, and for my company I have a list of acronyms that I'd like the Agent to refer to when employees ask questions because the same acronyms could have different meanings externally.

So I created a table within a word document where I put the acronym, full name, and description in front of it. Then, I created a topic for acronyms where in the Trigger I mention "This tool can handle queries like these: what does this acronym mean, explain this acronym, acronym definition, define this pattern acronym, pattern acronyms list, what does this stand for? meaning of this acronym? meaning? definition?" Finally I created a "Generative Answer" node where I explicitly only point to my word document containing the acronym list and I've turned off the "Allow the AI to use its own general knowledge". I set up the content moderation level to high and I've given it examples to learn from like this: "Find the phrase/acronym/abbreviation in the user's question and search for that phrase within the Energy Acronyms.docx line by line. For example, if the user asks "what is PPA?", "PPA?", "what does PPA stand for?", "PPA acronym?", "Definition of PPA". You should search for the phrase "PPA" in the word document line by line, if you find it you will return the full name and the definition mentioned in front of the phrase, and if you don't find the phrase, you will say "I didn't find the mentioned acronym as part of the company's acronym list!!!!""

However, sometimes, it reads the acronym from my document correctly and returns my company's definition of it and sometimes, it totally hallucinates, or is not able to read the acronym from my document. Is there any way I can improve its accuracy? Could you explain to me, how the word file is interpreted by Copilot?

Categories:

Topic creation & management