This channel will be a place for me to share my knowledge and experience around PowerShell scripting, Microsoft 365 and Microsoft Azure adventures as well as the occasional tech review.
I've been looking around for months for this exact use case. Thank you so much 👏. However, I still had to add title:[document name] in metadata for all documents in container (i did it manually) Few more things to consider: - using Basic tier for Azure Search service limits documents to 65536 characters, which in my case is 10 - 20 pages of pdf document. This is pretty restrictive. - when giving access to references there are additional settings necesarry on the blob container But most important - much better results comparing to connect Copilot to Sharepoint that has same documents in library.
@@markshark337 yes that is true about the Character limit. I am primarily using this for IT help desk tickets and documentation internally for my company, so those aren't that long. For large PDFs, you might need to break them up to stay on basic tier.
Good morning Dave, I'm trying to recreate the Chat environment on my personal Azure account using the free $200 service credit. When I get to the 'Add a data source' in the chat playground I follow your instructions for adding the 'Data source', and 'Data management' sections. When I get to the 'Data connection' section I have a choice to use 'System assigned managed identity' or 'API key'. Selecting the 'System assigned managed identity' creates an error. 'Failed to validate the selected authentication type. Please resolve the following errors and click next.' with the following two errors 'The Azure AI Search resource has disabled RBAC authentication' and 'The Azure OpenAI resource has disabled system assigned managed identity'. Do you know how to resolve these two errors. Or do I need to use the 'API key'? When I select the 'API key' the Chat bot produces a timeout not enough tokens error. Do you have any ideas or thoughts for resolving the errors? Thanks for your time!! ~David UPDATE: Boy, oh boy - this was a challenge to figure out!! I needed to change the Resource Group access control (IAM) in order for the for the managed identity to work. Sheesh!!
Hey! I tried to set it up with no success. How did you set up model data in the connection properties of the Azure OpenAI connection? Under "General" settings deployment name is clear, for the API version I tried both 2024-05-01-preview and 2023-06-01-preview but still getting "No information was found". The solution will be in connection properties.
I made a video reply to answer your question. Please see that here ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-vzLNfvC1K64.html If you have further questions, feel free to shoot me an email at david@davidjust.com
Man, you’re awesome. I wasn’t expecting such a quick answer, let alone in the format of an additional training video. I believe the issue in my case was with deploying from Azure OpenAI Studio to Copilot Studio Custom Environment, which is still in preview. I deleted the entire setup and first set up the Azure OpenAI connector from scratch in PowerApps. Then I added the Copilot configuration step by step, and it started to work. Many thanks for the explanation, I learned a lot!!
how to restrict copilot to invoke this APOD topic only if the user query is related to Nasa picture.. When i tested, i had a word document in the conversational booster + this APOD topic. and If the user ask the question lets say -when is the next snow removal day ?, the copilot is navigating via this APOD topic and giving me the answer we set there .... What am i doing wrong here .. I gave the description as -this should be used only for queries on NASA and Model description as --use this tool to get the NASA picture of the day
Ive ran into this as well. It appears that the conversational boosting topic takes the lowest priority and only triggers as a fallback. If you have any topics using generative AI, they may be triggered even though your topic description did not match the users query. To work around this, I recommend adding another topic with your uploaded documents as the knowledge source. Add a topic description to match the general theme / topic of those documents. I hope that makes sense, let me know if you need further clarification.
@@pwshisfun Yes, created two custom Topics like below 1. KB Article -- with uploaded documents as the knowledge source 2. APOD topic -- similar to this tutorial. So now when the asked the below question,it answered me from the APOD topic. which is wrong. In this case, it should NOT refer #1 or #2 above. User.Question : when is the next snow removal day ? Expected Answer - Sorry i dont know this on how to answer or similar..
Hi @Dave, I have an use case, lets say my data resides in the below three index. data_sources = { "Knowledge Base": "api.ourkba.search.windows.net/kba", "Press Releases": "api.pressreleases.search.windows.net/latest", "Closest Locations": "api.geolocation.search.windows.net/precincts" } Based on the user query, we need to get the response from the relevant API. What option we have in MCS? can we leverage Actions here ?Thoughts ?? eg : if user.query = "who is the new ceo of our company(XYZ)" -- MCS should make the call to "Press Releases" if user.query = "what is generative AI" -- MCS should make the call to "Knowlege Base" --
You need to enable generative ai under settings / generative AI. Where it says "How should your copilot decide how to respond?" Select generative (preview). This will create the conversation boosting topic.
@@pwshisfun ok thanks! Now the OpenAI can works on copilot studio, but only in the "test your copilot", I am unable to publish, no error in topic checker, coming back to your channel, you actually successfully published to teams!!! you did any configuration on azure? such as grant access etc?
@@ctvs-lee What happens when you try to publish? For teams, you need to publish the bot, then enable and configure the teams channel learn.microsoft.com/en-us/microsoft-copilot-studio/publication-fundamentals-publish-channels?tabs=teams
Thank you for sharing the knowledge Dave. For the past two days I've been struggling to figure out how to display SharePoint list data returned to Copilot by a Power Automate Flow. Your explanation has helped sort a few things out in my head and provided something new for me to try. Thank you!
Great video! Question on the Citations provided with the responses. Do you know if it's possible to go directly to the 'point of reference' in the documents? One of the docs is over a hundred pages and I'm thinking about breaking down the document into smaller chucks if I can't go directly to the reference point. Any ideas or work-arounds?
That is a good point. I think your idea of breaking up the document is the best bet. Azure AI search does not consider page numbers or anything to indicate the place in the document, so the citations will just have the document name and link to that document.
