In this repo there are 3 folders, representing 3 different AI phone call sessions. Our HR agent has called 3 different candidates and each have responded in their own unique way!
We want you to build an AI agent with either Python or Javascript, using your preferred AI provider, that evaluates call sessions and reports back whether or not the call was successful.
There are levels to this challenge and you can go up to any level as you want. It's totally ok. I would also recommend you do them one by one.
In this level, we only require the text transcript of each session, in whatever format you see fit. Write a script that generates transcripts from audio files.
Was the call "resolved"? We want a yes/no answer to this question. However, the exact definition of "resolved" is vague here, that's part of the challenge!
Ok, we now know whether the call was resolved or not, but we want to know more exactly how it went. Write an agent that scores the call out of 10 (you define the criteria), and provides a paragraph of summary, of what happened in the call, what went wrong or right.
Alright, now that we have the analysis and resolution status, we want to know what action should we take after the call. For example, should we call the candidate back? Should we escalate to a human? What should we do if the call went perfectly? For this level, your agent should generate a "Tool-call"-like JSON. You don't have to implement the tool itself. Just make up some name for it.
In the real world, we often don't know how far we can push a certain solution, or how we can make a "6 star experience" for our customers. In this level, you can challenge yourself to make your "Call Supervisor Agent" more useful, to whoever you imagine is going to use it.
Good Luck!
p.s. if you got stuck, don't give up. I'm always ready to help.