Exploristo: from points on a map to connections by meaning
My old project Exploristo started a long time ago quite simply. As it usually goes with projects like this: you solve some small problem of your own, you do one quite sensible thing, then it quietly pulls the next one along, and before long it’s too late to pretend this was ever a small experiment.
The original task was fairly down-to-earth: after moving to Germany, I had absolutely no idea what interesting things were around — both in larger cities like Düsseldorf and Cologne, and in small villages. The technical solution was fairly obvious: collect points from open sources like OpenStreetMap and Google Places, filtering for tourist attractions and medieval castles, put them in a database and see what comes out of it. Once there was enough data, the natural next step was to group points into clusters and build routes from them.
And in some sense, it worked. But it became clear fairly quickly that physical proximity on its own explains too little. Two places can stand right next to each other on a map and have absolutely nothing in common — for example, the ancient Roman gate Porta Nigra in Trier and Karl Marx’s house in the same city. The result was a cluster that was formally correct but semantically weak. And sometimes just strange.
It became obvious: the problem wasn’t in the clustering parameters. Without rich context, it can’t be solved at all.
What was needed wasn’t just an answer to the question “what is near what,” but also an answer to “what is connected to what, and why.” Because what people are usually interested in isn’t geometry as such. They’re interested in history, logic, theme, connection. Otherwise a route turns into a chain of points rather than something meaningful.
This is where Cogentis AI came in.
Open sources contain far more than coordinates and a set of tags. They hold context: historical events, architectural styles, eras, notable people, functional roles of places, and all kinds of other connections from which meaning gradually assembles itself.
When places start connecting through this kind of context, the grouping immediately becomes more interesting. A route stops being just a sequence of objects that happen to be within walking distance of each other. It acquires an internal logic. For example, “Marxism” or “ancient Rome” in that same Trier.
And the most important point here is that these connections aren’t pulled out of thin air by a language model at the moment of answering. They are grounded in sources that can be traced. This isn’t magic, and it isn’t an act of machine confidence with a knowing look. It’s simply careful work with knowledge.
That’s precisely why Exploristo turned out to be such a fitting real-world application for Cogentis AI.
Right now it still looks, in places, like a slice of Wikipedia — just arranged on the plate a little more neatly. Though there’s nothing surprising about that, since Wikipedia is indeed one of the sources.
The first thing a casual visitor will likely notice is the texts. They’re clunky, angular, and in places plainly wooden. At first glance this looks like a flaw. In practice, it’s one of the most important features of the whole construction. Because the text says only what the system actually knows. No more. If a fact is in the database, the system writes about it. If facts are contradictory or inaccurate, it says so honestly. If it doesn’t know something, it doesn’t make it up. Without those favourite tricks of modern chatbots, where a phrase sounds convincing simply because it’s well glued together — not because there’s anything behind it.
If you ask a regular AI chatbot to describe any of these places, it will almost certainly answer with more, livelier, more fluently and more impressively. But quite possibly less reliably. The text will be more pleasant, and the line between fact and invention will be blurrier. And it will be nearly impossible to understand where any given thought came from in the first place.
That’s why the value of Exploristo right now isn’t that it’s yet another personal travel project with place cards and routes. Its value is that it’s a tangible demonstration of a different approach to AI systems: not trying to speak about the world as convincingly as possible, but trying to represent knowledge about the world in a form that can be verified, traced back to sources, and trusted within reasonable limits.