I was driving from Toronto to Ottawa one day and I was bored. I had decided to take Highway 7 instead of the 401 just because the scenery was a little bit more interesting. As I passed through small towns and villages, I found myself wondering, “What happens in this town? What am I missing by just blowing right by it?”. Wouldn’t it be cool, I thought, if I could be playing a podcast that talked about each of these places as I drove by them?
When I got back home, I started on this proof-of-concept app, which I call Conversational. Conversational is a Rails app with LLM-backed agents having a podcast-style conversation about the significant things close to your current location. As you move around, your browser sends periodic location updates to the app’s server. The app then uses Google Places API to determine what is notable and close by, then uses Wikipedia and ChatGPT to generate a plausible back-and-forth conversation between two “hosts”. There is a third (silent) agent in the background that acts as the program manager, keeping the other two agents on topic.
It needs a fair bit more polish, and there’s no doubt you’re listening to AI, but its a promising idea. I’m particularly excited about advances in text-to-speech, like what Eleven Labs is doing. As these models get better, apps like this will only become more convincing.