Making public data truly accessible

Bringing regional data into one place

Our collaboration with Regione Toscana began in 2022 with the design of Data Monitor, a platform bringing together the data of the Osservatorio Sociale Regionale.

We designed a system where social and healthcare indicators could be explored through maps, charts and tables, filtering information by topic and geography: from a regional overview down to individual municipalities.

For the first time, a vast amount of fragmented information became available through a single interface, creating a shared tool for researchers, policymakers, local administrations and citizens.

Accessible data, but accessible to whom?

Data Monitor made the data available. But over time, a different question emerged: available to whom?

The richness of the platform was also its main barrier. A single search could involve multiple fields, indicators, geographical levels and filters. For researchers and expert users, this depth was essential. For mayors, local administrators or citizens looking for a specific answer, even formulating the right query could become a challenge.

We didn't want to reduce the complexity of the data or remove possibilities from the platform. We needed to make that complexity easier to navigate.

What if you could simply ask the data?

This led us to explore AI not as a way to generate answers, but as a new interface to the existing system.

After several months of development, we designed a dedicated LLM-powered search agent that interprets questions expressed in natural language and translates them into structured queries, applying the fields and filters required by Data Monitor. The agent does not calculate results, generate interpretations of the data or rely on RAG to produce an answer. The platform remains responsible for retrieving and displaying the underlying information.

The LLM does one thing: it understands what the user is looking for and turns that intent into a query the system can understand.

In this way, AI becomes a bridge between human questions and structured public data. The complexity remains, but users no longer need to learn it before they can ask a question.