Close Call, a Voice AI Sales Coach
TL;DR
I built an open-source voice agent that lets salespeople practice tough calls against an AI customer. It runs on Pipecat + Gemini 2.5 Flash, costs about $0.14 per call, and my team's first reaction was genuinely enthusiastic. The code is on GitHub if you want to try it.
"Sales cures all ills" is a phrase one of my first mentors repeated often, and it stuck with me. Before I got serious about software development and before meddling with AI engineering, I was a salesperson. I still miss the thrill of closing a deal and outmaneuvering your opponent. It was fun. Now that I'm in a position to use AI at my company, I came up with an idea: use voice-enabled AI to help train and coach our salespeople.
From a technology standpoint, this was a relatively simple implementation. I built Close Call, an open-source voice agent where a salesperson picks a scenario, clicks "Start Call," and talks to an AI customer who pushes back with realistic objections. Immediately after, the system performs an assessment and provides feedback that highlights where the salesperson performed well and what could be improved.
I've done an initial demo at my company. The feedback was positive, very positive, in fact.
The flow is surprisingly realistic. You're having an actual conversation with the AI, and it responds in real time. The scenarios are simple to set up. You can see examples in the repo. This is infinitely extensible: different approaches, different tones, different moods, different levels of pushback. In its current state, it's already generating real value.
The current feedback by the AI is still mechanical. It rates how you handled and maneuvered past objections, whether you stayed on track. In future iterations, I want to evaluate emotion and tone during the call. And beyond that, video analysis of posture and physical presence. You might not believe that physical disposition matters, but it comes through on phone calls too. There's a lot of room to grow from this starting point.
The Stack
Close Call runs on Pipecat with Gemini 2.5 Flash Native Audio for native speech-to-speech at sub-second latency. A 10-minute practice call costs roughly $0.14–0.40. Post-call coaching feedback is also generated by Gemini 2.5 Flash, from the transcript. The whole thing is Docker-packaged for easy sharing or can even be run as a desktop application.
I wanted to share this because it's an example of a very simple application made possible by recent model releases. Gemini's live model came out about a week ago. It's not the first model that could handle live conversation, but it was cost-effective enough to finally make me invest time in an idea I'd been sitting on.
This is what understanding the AI landscape looks like in practice. You don't have to orchestrate a complicated business process to deliver value. Sometimes the low-hanging fruit delivers a big punch. Or, as my mentor would say, cures all ills.