What if a large language model could not only answer questions, but also search for up-to-date data, download a map, and calculate whatever you need to know about a specific location? This is exactly the idea developed in the master’s thesis of Ing. Matúš Botek, a student of the Faculty of Information Technology at the Czech Technical University in Prague (FIT CTU). The result is GeoChat Assistant – a smart assistant that connects the capabilities of artificial intelligence with geographic tools. For his work, he received the Dean’s Award in the summer semester of 2024/2025.
A GeoChat Assistant user simply selects the area they are interested in, and the Assistant decides what is needed – it can download a map from public services, retrieve population or weather data, or even perform terrain analysis. Thanks to this, its answers are not only easy to understand but, more importantly, based on current and real data.
“I wanted to show that language models can be extended with tools that allow them to solve tasks beyond purely textual domains. I chose geographic data to demonstrate that combining a model with external tools can improve its ability to interpret and process complex real-world data,” says Matúš.
GeoChat Assistant is designed to autonomously choose the right tool and combine the results into a coherent response. In practice, this means that it does not behave like a passive ‘answering machine’ but actively plans and decides how to solve the task.
The system was tested both on a set of predefined tasks and in collaboration with experts in geoinformatics. Their feedback confirmed that the assistant can work more accurately and reliably than standard large language models that rely only on their learned knowledge.
“For me, feedback from experts who work directly with geodata was extremely valuable. They appreciated the intuitive interaction with GeoChat and its ability to simplify access to obtaining and interpreting geographic data,” adds Matúš.
The thesis also outlines directions for further development of such a system – for example, improved visualization of results, integration of additional data sources, or more precise tool usage.
“Since submitting my thesis, I have continued to add several new data sources and experimented with creating new tools. Although I am not currently actively working on the GeoChat assistant project, I now use the experience gained during its development in other projects focused on using language models as agents with access to various external tools,” clarifies Matúš.
“The results of the work are very interesting, useful, and easily extensible. A significant part of the work has been incorporated into the PoliRuralPlus Horizon project and submitted for presentation at the FedCSIS 2025 scientific conference,” says the thesis supervisor, Mgr. Alexander Kovalenko, Ph.D.
