Abstract

Imagine, what would it look like to be able to ask advanced analytic questions to your urban data, like talking to a human expert, without having to write any code? Latest versions of ChatGPT enable that to some extent, but are quite limited in terms of the types of provided analytics (mostly standard data summarization), and its reliability, as well as the size of the data they can handle. Our team has developed an advanced prompt engineering framework for modern Generative AI libraries, tailored to urban data of spatio-temporal networked structure, allowing to introduce available urban datasets, along with available advanced spatio-temporal network analytics tools for them, and implement a feedback loop for autocorrection of the code provided by GenAI, making the final answer much more elaborate and reliable, compared to using plain ChatGPT.

We evaluated this framework on several pilot datasets, and are now inviting you to join us on an exciting journey of introducing many more datasets to the system from NYC Open Data, and other open data sources on various urban activities across the world. While experimenting with the new datasets, you will also further advance the framework itself, enabling it to support even more types of advanced urban analytics in a more reliable way. Furthermore, we shall experiment with custom GenAI solutions for data analytics, as an alternative to relying on OpenAI or other generic APIs. Completion of the project shall open up the potential of urban data analytics to broad non-technical audience, and also contribute towards the state-of-the-art of generative AI for data analytics.