The HARMONIA project aims to develop methods for using Large Language Models (LLMs) enhanced with knowledge derived from behavioral and statistical data to support informed and inclusive decision-making in public policy processes.

The project focuses on creating trustworthy AI to support political choices—from listening and decision-making to communication—guiding both citizens and authorities toward new behaviors.

While LLMs offer powerful capabilities in natural language understanding, they remain unreliable for public decision-making. Their responses can be generic, biased, or even incorrect—especially when dealing with underrepresented communities or unfamiliar situations. This limits their usefulness for local authorities who need to understand the diverse needs of real people.

HARMONIA addresses these challenges by enhancing LLMs with structured semantic knowledge drawn from real-world behavioral and statistical data. Through neuro-symbolic approaches—combining ontology-based models with advanced AI techniques like

Retrieval-Augmented Generation (RAG) and Knowledge Injection—the project is designing models that can reason with context and take on multiple perspectives. The goal: more inclusive, transparent, and trustworthy AI tools that strengthen the relationship between citizens and public institutions.