Note: This write up is not yet complete.

I used the openAI gpt4 and whisper API’s to create a simple ‘tour guide’ allowing a user to learn more about the history and culture of their or any location. This project is simply a proof of concept for an interesting use case of language models meant to deepen our connection to our surroundings.

I had this idea a few years ago while I was going for a walk in Lower Manhattan. I was curious about the ignored history that sat resting under the cobble roads and the stone industrial walls and started thinking about how cool it would be to have a tour guide for my everyday adventures. A way to dynamically learn the seen and unseen history - to see the present as I hear the past. At this point GPT-3 had just come out, and it became clear that with the right data, writing a localized script to learn about where you stand was soon going to be possible.

About a year or two later OpenAI released their text-voice model, Whisper. There were no longer any strict reasons my idea could not be achieved, so I figured I might as well try and see if I could make myself a simple ‘Tour Guide Assistant’ GPT-4 wrapper. I had no intention of making this into a convenient and polished piece of software, as much as I resonated with the idea, I suspected that like any other LLM wrapper, this will become native within some AI system baked into most smartphones within the next couple of years. With this in mind, this project is really just proof of concept, and was scrapped together in Jupyter Notebook. What I learned from this project is a comfort working with API’s, and an understanding of the kinds of data that is accessible to developers. Another takeaway is that a company with sufficient information could create an excellent interactive tour guide, this could be an excellent expansive application of artificial intelligence if approached correctly.

I wanted there to be four options for the tour guide:

• Free Roam - uses position data to talk about relative surroundings as you go Walking around a city Walking around a park Walking around town
• Path - uses a path to generate an optimal historical story to tell Walking to a destination in a city Driving following some directions
• Suggestions - Considers location, time of day, habits, and recommends nearby activities, food, and places to explore
• Explorer - Essentially a navigable globe (like Google Maps) where we can click somewhere and the model will tell us about what it is like to be there right now.

These are the four options I think should be on a tour guide type of system. I focused only on the ‘Explorer’ option to keep things simple.

[Write Up Not Yet Complete]