Objectives
The project aims to develop an LLM-based conversational query system for relational databases, enabling users to access data through natural language and overcoming the limitations of traditional database querying methods.
A central objective is to enhance the LEDA platform by introducing new ways of accessing and consulting data, reducing the technical complexity of queries and improving the overall user experience. The project therefore includes the development of an operational prototype with dedicated interfaces for both end users and system administrators, designed to support experimentation and validation activities.
Particular attention is also given to system robustness and reliability. The solution is tested against real and complex scenarios, including out-of-context requests, potential bias and edge cases, in order to assess the model’s behaviour under non-ideal conditions. Finally, the project aims to establish a scalable and reusable technological foundation that can be adapted to different business and organisational settings, paving the way for future extensions and applications in other domains.
Initial Challenge
The project’s main challenge was to overcome the limitations of the existing LEDA management system, which required specific technical expertise to query data and presented a steep learning curve for non-technical users.
The key issues included limited accessibility to structured data, insufficient flexibility in database querying, and the need to ensure the quality and reliability of the system’s responses while preventing interpretation errors and incorrect queries.
Solution
The solution was developed through a collaborative and modular approach, culminating in a Proof of Concept that integrates LEDA with a Large Language Model for conversational data querying.
The system implements a multi-agent pipeline that includes:
- smart entity normalisation to improve the semantic understanding of user requests;
- intent detection and query reformulation;
- automatic generation and execution of SQL queries, supported by self-correction mechanisms;
- orchestration of the different processing modules;
- generation of both narrative outputs in natural language and structured results.
A pilot web application was developed to support experimentation, allowing users to test the overall experience while enabling administrators to manage the system. The project concluded with a structured phase of functional and adversarial testing designed to validate the system’s reliability under both real-world and non-ideal conditions.
Benefits
The Proof of Concept represents a significant step towards more advanced and inclusive use of corporate data.
Its main benefits include:
- faster and more intuitive access to data through natural language;
- reduced SQL query complexity for end users;
- improved response quality and robustness through normalisation and control mechanisms;
- increased operational efficiency in the use of information stored within the management platform;
- creation of a scalable technological foundation that can be adapted to different application domains and organisational contexts.
The project demonstrates how the integration of traditional management systems with advanced artificial intelligence technologies can enable new ways of interacting with data, improve decision-making and unlock greater value from existing information assets.
IFAB’s Role
IFAB played three complementary roles in the project: funder, operational coordinator and technical partner. As a funder, IFAB selected the project as part of its investment programmes in research, innovation and technology transfer, recognising its scientific and practical value and enabling the activities carried out by the technical partners. From an operational perspective, IFAB assumed overall responsibility for project management and administration. Its activities included financial reporting and budget monitoring, oversight of partner relationships, the organisation and coordination of operational meetings, and the collection of project results through regular progress monitoring.
IFAB also worked alongside the start-up Ada, which was responsible for the technical development of the software, creating the operational and collaborative conditions required for the project to progress consistently and in line with its defined objectives.
IFAB’s technical contribution focused on three complementary areas:
- designing and providing a dedicated data lake infrastructure for social and personal services, capable of hosting, integrating and managing heterogeneous data from multiple sources and serving as the foundation for testing and validating the AI solutions developed during the project;
- contributing AI and machine learning expertise for data monitoring and the identification of patterns and trends, supporting both the platform’s operational management and analytical activities;
- developing and testing Large Language Models to enable natural-language interactions within the platform and make access to information more immediate for third-sector professionals.
Participation in the LEDA project enabled Fondazione IFAB to consolidate cross-disciplinary expertise at the intersection of artificial intelligence and social innovation, further confirming its mission as an accelerator of high-impact initiatives for the local area and the wider production ecosystem.
For further information, please contact: projects@ifabfoundation.org
