I had a prompt that could return details from a Dataverse table that was working extremely well until recently. The prompt essentially searches a table for a user id then returns details about them.. Or it used to.
Unfortunately not any more. I'm fairly confident the reason is that its only looking at a small subset of rows in the table now (even though there's only a few hundred rows in total) - Setting a hard coded value in the filter attribute also proves this, but that isn't an option as part of the goodness of the prompt was extracting the user ID from the input text.
Its almost like I need to chain 2 AI Builder prompts together (The first to get the user id) but that seems a waste, especially being it previously worked. I'm hoping someone may have better ideas? In case it helps here's the prompt I'm using:
Given the conversation text, identify the three-character User ID (e.g., REDACTED). Search for this User ID in the <users> Dataverse table . Once found, respond to the user's question, ensuring you repeat the user ID they asked about the state the journal poster's <full name> and <email>. Ensure the full name is in **bold** and format the email as a clickable link. If you couldn't find the journal poster, thats fine, just let them know.
If you successfully found the journal poster respond in a friendly tone. If you couldn't find the journal poster, stick with a polite and professional tone and ask them to double check they entered the Sun user ID correctly.
-----
Here is the conversation text:
