Hi,
We're working on creating a chat bot to answer questions related to medical insurance documents. We've set up the necessary resources in Microsoft Azure (Azure AI Search index, a semantic configuration and a GPT 4o deployment) and we tested the bot on the document in the Azure OpenAI Studio Chat Playground. It answers most questions quite well.
Now, we'd like to use the same resources to Copilot Studio. We've created a Copilot and updated its boosting topic with a connection to OpenAI, using the same LLM parameters (same OpenAI deployment, temperature, top p, max tokens), the same AI Search Index (same content, title and file name fields), the same semantic search configuration, same system message and AI Search Index settings (strictness, top n documents, etc).
When we ask questions to this bot, it does answer, but it fail to find an answer to some of the same questions that the chat playground does not fail to answer. As far as we can see, all of the settings in the boosting topic are the same as the playground, aside from the API version (Copilot Studio only permits up to 2024-02-15-preview for the OpenAI API, where the chat playground is on 2024-05-01-preview).
Are we missing something?
12
Share
Report
All responses (
Answers (
