Hi, I’m Takahashi.
I’m working on the AI behind timespace ZINE, which creates articles about places from photos taken around a city. Take a photo of somewhere that catches your attention, and you get a story about what’s behind it.
This time, I’d like to share something I found interesting while using the app in Linz.
Even when you’re helping build something, using it yourself can still surprise you.
From a Shop in a Photo to the People Behind It
What stood out to me most was how far the research could go, starting from a photo.
It uncovered detailed information about the shop in the image, even reaching the people connected to it. I remember thinking, “You can get all that from a photo?”
When you’re walking around a city, you sometimes notice a shop that interests you. But how often do you look into it far enough to learn about the people behind it?
Stories I would have walked past without ever looking into them were turning up in the app.
That was what made it interesting to use.
What Goal Do We Give the AI?
One part of this system I find particularly interesting is the goal we give the AI.
When you want information about a place, asking it to “research this place” seems natural.
Our approach is to set the goal as what we ultimately want to create—an interesting article—and gather the information found along the way as material we can use later.
What do you need to know to write an interesting article? Is a basic overview of the shop enough? Would it help to look into the people connected to it, or the history of how it came to be?
Having an article to work toward gives the research direction. The information gathered in that process also becomes material for our database.
When deciding what to ask the AI to do, we set its goal with the final outcome in mind. I think that approach is working well.
The Research That Makes the Story Interesting
Using the app in Linz also made me realize how much the quality of the articles depends on the research done beforehand.
We research places using photos in advance and prepare the material. When someone scans a place with the app, it uses that material to put an article together.
“Reassembling” is probably a good way to describe what the app is doing at this stage.
Because that research has already been done in depth, taking a single photo can lead you to a story you would never have searched for yourself. In Linz, I got to experience the appeal of that process as a user.
But that also raises another question.
Since people are taking photos right there, on the spot, could we make better use of what those photos capture in that moment? Even at the same place, could we offer a different article depending on who took the photo and what caught their attention?
I’ll come back to that question—what article to offer the person taking a photo, here and now—in another post.
