A week ago, I was in a job interview, and we discussed Agents. A long time ago, when Power Virtual Agents was only topic-solution oriented, I remember seeing a virtual agent that used to match your answers to a certain beer. So I thought ok, let's do something with Coffee, I decided to call it Coffit, as it is a combination of Coffee and Fit. Yeah, a bit lazy I know, but it sort of makes sense.
I started giving some of my inputs into the initial page of Copilot Studio, where I can actually prompt the idea. Turns out I wasn't really happy with the results, so I needed to add some logic behind with my own hands.
Let me guide you on the process I made. Below you will find an image of the overview of my little Coffit (in the left image you can see the little logo I made for the agent).
In the overview, there are some remains of the prompt piece that I initially started, and then decided to follow up manually. See the instructions, those are prompted.
For knowledge I added specific resources that had an intimate relationship with coffee, it goes without saying that some websites contained good info about it, classical coffee brands and a really interesting resource that was very descriptive on types of coffees.
I created a topic to help me design the questions and answers and take those variables into an AI prompt. So in the video below you can see how the four questions have three predefined answers. Each of the answers is stored in a variable.
So the four variables feed an AI prompt that retrieves the best fit based on the four components:
🌄 adventure
🏋️ challenge
🤙 companion
💥 excitement
Once these variables are collected, the prompt is triggered and the user receives a message with the perfect coffee match. The outcome can be found at the end of the image on the right. This will be displayed as a message that not only contains the type of coffee but also gives a cool idea of why that is the perfect fit, meaning the reason why it was a perfect fit.
AI prompts are part of the Power Platform cool stuff that can be retrieved into apps, automation and agents, and that is exactly what we are doing here. The prompt below is not perfect nor ideal - I was lazy with my writing but it does the job and that is all I care about now.
In the video below you can see the agent in action, I'm pleased with the final solution and would like to add some ordering actions directly from the agent. I'm considering adding voice features so it's more inclusive and adapts to different scenarios and needs.
Find below a small Coffit demo:
This solution - silly as it is - can illustrate so many cool scenarios where it can be applied. I'm thinking cultural fit company candidate, fit between open roles and candidates. Probably some other scenarios that are not only HR ones, can you think of any?
Microsoft resources:
Comments