Best of both

[teledactyl]

I've developed a nice Bot using 'sequence' augmentation that can automate several things for me in my stage-view app within Teams.
I also want it to act as an experienced engineer so it can begin to give advice, so I've used Vectra to encode a few documents on a fairly narrow engineering discipline. When using 'none' as augmentation, I get some good, detailed, results.
But, when switching back to 'sequence', the advice becomes quite terse.

In VS code's terminal I can see that the prompt is picking up similar content, but the results are consistently different.
How can we get the best of both worlds?

[lilyydu]

As per live discussion, additional samples to consider for encoding additional docs/information

https://github.com/microsoft/teams-ai/tree/main/js/samples/04.ai-apps/h.datasource-azureOpenAI
https://github.com/microsoft/teams-ai/tree/main/js/samples/04.ai-apps/g.datasource-azureAISearch

[Carter425]

Here is the form to apply to Azure OpenAI service

https://customervoice.microsoft.com/Pages/ResponsePage.aspx?id=v4j5cvGGr0GRqy180BHbR7en2Ais5pxKtso_Pz4b1_xUNTZBNzRKNlVQSFhZMU9aV09EVzYxWFdORCQlQCN0PWcu

[Carter425]

Here is the sample for Azure OpenAI On Your Data https://github.com/microsoft/teams-ai/tree/main/js/samples/04.ai-apps/h.datasource-azureOpenAI

[Carter425]

Here is a nice video walkthrough to go along with the tutorial: https://www.youtube.com/watch?v=VLP32Ci02M0

[singhk97]

Hi @teledactyl, you can also try adding a few examples of user/model interaction within the prompt. This technique is called few-shot prompting, and it's really effective in guiding the model to return responses in a specific format.

• https://www.promptingguide.ai/techniques/fewshot

[teledactyl]

I love this prompting guide: it's a real eye-opener.