CHOOSEAgents & automation
Jev-Powered Sales Script Assistant
Kalen Jordan@kalenjordan
1:42This project uses Jev to analyze call transcripts and guide users through a decision-tree-based sales script, displaying the next question to ask.
I got access to Jev and was planning to spend the whole weekend building something with it. By the end of last night I'd pretty much decided it was useless.
I was looking through use cases on Twitter and couldn't find anything I wanted to do. The closest idea I had was something with structured product data and Shopify's UCP, but Shopify's probably gonna do that themselves. They kind of already are.
So I gave up and went back to something I'd been working on: cold calling.
I'd started with voice agents, but people immediately realized they were talking to AI and it wasn't working. So I figured I'd just do the calls myself.
Naturally, I managed to turn that into a software project. Auto dialing, navigating voicemail systems, all the stuff around actually talking to someone. Building it was fun. Getting myself to make the calls was like pulling teeth. But now I'd spent all this time building the thing, so I kind of had to use it.
What I wanted was for it to run in the background while I worked in Codex. Get through the voicemails, the press ones, the "say your name and we'll try to connect you." If somebody picks up every five or ten or fifteen minutes, that's fine. I'm working anyway.
And I was like, oh. That would be a good use case for Jev.
So I wired it in to detect what's on the other end: a phone system, a voicemail, a voice agent, or a live person. I've actually been running into more voice agents answering the phones lately, which is interesting given where I started with this.
Then there was the call script. I'd been thinking through questions with ChatGPT and keeping a notepad open, but once you're on the call you get flustered and forget what you were gonna say.
So I built a decision tree for the sales script. Jev checks the current transcript against that tree, figures out where we are in the conversation, and puts the next question front and center on the screen. You can just glance over and see what to ask next.
The funny thing is that by the time I built it, I'd spent so much time thinking through the script that I pretty much had it memorized. I didn't need it as much anymore.
But now I'm wondering if that same structure could make the voice agent work better. Give it the decision tree and a really fast text-to-voice API, instead of asking the model to figure out the whole call as it goes.
I went from planning a whole weekend around Jev, to writing it off, to finding two things I wanted it for as soon as I went back to working on something else.
