What is it about?
AI assistants are usually trained on information from the past, so they may not know what is happening in a city right now. This work shows how an AI assistant can be connected to live city data and receive only the information that is relevant to a specific place and moment. We tested the approach in Madrid with examples involving tourist attractions, traffic lights, and street lighting. The results show that selecting data by location and time can help AI give more useful and accurate answers while reducing the amount of information it must examine. The study also identifies current limitations, especially when AI systems must process very large amounts of data or respond quickly.
Featured Image
Photo by unavailable parts on Unsplash
Why is it important?
Cities generate large amounts of information that changes continuously, but LLMs normally rely on knowledge captured during training. This work provides a practical way to connect them to current city data while selecting only the information relevant to a particular place and time. Its importance is especially timely as cities increasingly adopt digital twins and generative AI. By using interoperable, open technologies, the approach can be integrated with different data sources and AI models rather than tied to a single provider. It could help cities build more accurate and useful assistants for residents, visitors, and operators, while reducing unnecessary data processing and revealing where current LLMs still struggle with scale and response time.
Perspectives
What I find most compelling about this work is that better results do not always require a larger or more expensive AI model. A carefully designed approach can make an LLM far more effective by giving it only the right information, for the right place, at the right time. This reduces hallucinations and computing time while improving accuracy and response quality. For me, the real value of the work is that it addresses a practical problem: connecting AI systems to constantly changing real-world information in a reliable and efficient way. I hope this approach helps move GenAI beyond demonstrations and toward useful applications that can operate in real cities and solve real problems.
David Nazareno CAMPO
Universidad Politecnica de Madrid
Read the Original
This page is a summary of: Real-time Spatial Retrieval Augmented Generation for Urban Environments, ACM Transactions on Intelligent Systems and Technology, July 2026, ACM (Association for Computing Machinery),
DOI: 10.1145/3831675.
You can read the full text:
Contributors
The following have contributed to this page
