What question did this study set out to answer?

The central aim is to create a compliance framework that enhances transparency and trust in fertilizer regulations.

February 12, 20260 citationsOpen Access

Leveraging LLMs for fertilizer transparency in sustainable food systems

GGGagan Manjunatha Gowda ABAnsgar BernardiUniversity of Kaiserslautern AKAhmad Kadi

Puntos clave

The central aim is to create a compliance framework that enhances transparency and trust in fertilizer regulations.
Developed a hybrid compliance framework combining deterministic rule validation and retrieval-augmented generation.
Captured organization-specific activity logs in semantically structured JSON-LD linked to domain ontologies.
Performed deterministic checks and logical mapping of product function categories.
Utilized a retrieval-based language model to interpret compliance statements against EU regulations.
Initial tests indicate the hybrid approach provides more stable results than pure generative models.
The model generates weighted confidence scores and pathways for citable evidence.
Integrating compliance results into the Digital Product Passport enhances regulatory trust in the food sector.

Resumen

Ensuring transparency and regulatory trust in agricultural and food value chains requires mechanisms that not only document product compliance within the framework of the Digital Product Passport (DPP) but also make it verifiable. This paper presents a hybrid compliance framework developed within the context of the SUSKULT project, combining deterministic rule validation with Retrieval Augmented Generation (RAG) to assess the compliance of fertilizer products according to Regulation (EU) 2019/1009. Organization-specific activity logs are captured in semantically structured JSON-LD, linked to domain ontologies, and processed in a two-tier architecture. The first tier performs deterministic checks, including verification of labeling according to Annex III, and logical mapping of product function categories (PFCs). The second level utilizes a retrieval-based language model that interprets these statements within the context of relevant EU regulations. Initial evaluative tests with synthetic data demonstrate that the hybrid approach delivers more stable and explainable results than purely generative models by generating weighted confidence scores and citable evidence pathways. Integrating the compliance results into the Digital Product Passport creates a verifiable anchor of trust, serving as proof of regulatory conformity and enabling data-driven, transparent, and sustainable value creation in the food and fertilizer sectors.

Preguntar a la IA

Me gusta

Guardar

Ver artículo completo