N. 1071 - Can hybrid semantic-graph RAG bridge retrieval and reasoning in Economic Bulletin drafting?

Questioni di Economia e Finanza (Occasional papers)
by Giuseppe Bruno
October 2026
Vai alla versione italiana Site Search

Within the framework of AI tool evaluation, this paper proposes an advanced retrieval-augmented generation (RAG) pipeline to automate the drafting of Banca d'Italia's Economic Bulletins, with economist oversight at every intermediate stage of data collection, draft review and verification of tables and charts - the key elements of communication on the state of the Italian economy in the Bulletins.

The proposed approach enables AI-generated statements to be linked to their original sources with a high degree of reliability. At the same time, it can generate texts grounded in empirical evidence and economic theory, as reflected in previous editions of the Economic Bulletin. The role of subject-matter experts nevertheless remains essential in order to mitigate the risk of errors in AI-generated statements.

Full text

  1. 13 October 2026
    No. 1071 - Can hybrid semantic-graph RAG bridge retrieval and reasoning in Economic Bulletin drafting?
    PDF 1 MB

Sezione di approfondimento