Conference Agenda

Agricultural knowledge serves as the cornerstone of food production and decision-making. The information of agricultural knowledge is often fragmented across various publications from agricultural extension stations. Traditionally, such knowledge was acquired by querying extension experts; however, their demanding schedules make timely responses challenging. Furthermore, agricultural practices are highly region-specific, meaning that general information found on the internet may not be applicable to all local contexts. While recent advances in large language models (LLMs) offer a promising path for developing agricultural question-answering (QA) systems, there is currently no robust mechanism to evaluate the accuracy and quality of their outputs. To address this gap, this study proposes the creation of a benchmark in traditional Chinese specifically designed to evaluate agricultural knowledge QA systems within the context of Taiwan. The benchmark dataset spans six crop varieties and five distinct domains, featuring responses generated by experts to serve as ground truth. Using this benchmark, the performance of open-source LLMs with and without the integration of retrieval-augmented generation was evaluated. The concordance of LLM-generated answers was evaluated and scored based on the ground truth. This benchmark establishes a comprehensive framework for verifying the output of LLMs, ensuring their alignment with authoritative agricultural knowledge in Taiwan.

Cooperative Extension in the United States, administered through land-grant universities, operates as a federal–state–county partnership that translates scientific research into public knowledge and practical application. In today’s digital economy, where universities function as engines of knowledge creation, scalable and audience-specific dissemination of research-based information is increasingly critical. Extension professionals require integrated technological solutions that enable clientele engagement and deliver personalized, up-to-date, relevant content through their clients' preferred communication channels. This paper presents an integrated digital engagement architecture developed at the University of Florida’s Institute of Food and Agricultural Sciences (UF/IFAS). The system employs a distributed subscription management framework supported by a centralized database that aggregates clientele data from authorized websites through a secure subscription API. By eliminating data silos and standardizing data governance, architecture enables a unified platform and enterprise-level engagement management. The platform supports coordinated and automated dissemination of science-based content across multiple channels, including email, web, SMS, and social media. Results demonstrate improved engagement performance and communication efficiency through targeted delivery and integrated analytics. Future enhancements will incorporate AI to streamline content generation, targeted clientele engagement, and knowledge delivery.

The effective implementation of sustainable precision viticulture (SPV) necessitates a paradigm shift in management, requiring a multidisciplinary skill set spanning technical, ecological, and digital domains. Within the framework of the Erasmus+ project VTskills (ID: 101139985), an e-learning platform was developed focusing on four thematic areas: SPV overview, resilience, green, and Digital Skills. The platform offered distinct curricula tailored for higher education (HEI) and vocational education and training (VET) students. To evaluate learning quality and identify e-learning constraints, a validated 40-item questionnaire was administered to 198 students. Data were analysed using cross-variable and cluster analysis to compare perceptions between HEI and VET cohorts. Results indicated a high level of interest in SPV topics, which are deemed strategic for the modern viticultural sector. While the course structure was well-received, linguistic barriers—specifically the use of English in some modules—negatively impacted student appreciation. The findings conclude that while digital platforms are effective for knowledge sharing, robust professional preparedness in SPV requires an integrated approach that blends digital delivery with structured practical experience to bridge the gap between theory and field application.

As global food demand rises and climate change exacerbates resource depletion, traditional agricultural methodologies pose significant risks to European food security. While Agricultural Deep Tech including Artificial Intelligence, IoT, and Blockchain, offers critical opportunities for productivity enhancement and waste reduction, technological development currently outpaces its practical application. Consequently, European agriculture faces a structural "triple challenge": a technology adoption gap, a shortage of digital skills, and misaligned educational systems. To address this paradigm shift, this paper proposes the standardized profile of the AgriTech Manager (ATM) to bridge the divide between technological innovation, business operations, and sustainability. The ATM framework is supported by a Learning Ecosystem that connects Vocational Education and Training (VET) with Higher Education Institutions (HEI) through modular and Plan-Do-Check-Act (PDCA) based methodology. To validate the practical necessity of these competencies, the study presents an applied Deep Tech approach to water resource management, comparing water consumption metrics across a highly sensorized hydroponic prototype, standard greenhouse environments, and the FAO water consumption model. Ultimately, this framework provides an actionable blueprint for developing a resilient, digitally proficient agri-food workforce.

The expansion of Precision Livestock Farming (PLF) technologies has increased the need for graduates skilled in sensor-based data collection, modeling, and application. Recent surveys identified hands-on learning as a gap in PLF education. This abstract describes a laboratory module guiding students through a PLF data pipeline: sensor placement, data collection, cleaning/preprocessing, model development, and deployment. Using personal smartphones and the Phyphox app, students collected 2-second bursts of data during predefined activities (standing, sitting, walking, jumping). Three structured datasets emphasized replication and consistency: a single individual performing 30 repetitions per activity (Dataset 1), five individuals performing 10 repetitions per activity (Dataset 2), and two individuals performing four repetitions for validation (Dataset 3). Students augmented Dataset 2 with their own recordings to increase variability across individuals. Raw signals were converted into statistical features and used to develop classification models (WEKA), with Datasets 1 and 2 used for training and Dataset 3 for validation. Finally, students collected two minutes of continuous, randomly sequenced activities (Dataset 4) and tested their models for real-time prediction. Students completed the workflow, achieving classification accuracies of 74-99%. This activity provides a low-cost framework for teaching core PLF competencies while reinforcing experimental design and model robustness in PLF systems.

The use of autonomous machines in agriculture requires effective collaboration between human supervisors and autonomous systems. Safe and reliable supervision depends on shared situation awareness (SSA), defined as the degree to which both the human and the autonomous system perceive and understand the same operational state. Currently, no standardized method exists to quantitatively measure SSA in agricultural human–autonomous teams. This study develops and tests a novel assessment framework using a FarmBot platform as a case study. Twenty-five participants remotely supervise the FarmBot as it performs seeding, watering, weed disruption, and plant imaging tasks. During four structured supervision sessions, situation-based pop-up queries appear at random intervals. Participants respond based on their understanding of the system’s current state, while corresponding machine responses are generated using the FarmBot’s internal data through its REST API. Human and machine responses are compared to quantify alignment. Experimental testing is ongoing to examine how effectively the proposed approach captures shared situation awareness during autonomous task supervision. This work establishes a structured method for evaluating shared situation awareness in autonomous agricultural systems, supporting improved interface design and safer human–autonomous collaboration.

Project TWIN-IN is an Erasmus+ initiative dedicated to facilitating twin (green and digital) transitions by developing, testing, and validating a new model for education and training, called “Twin Transition Schools”. Exploiting four competency frameworks developed by the EU (DigComp, GreenComp, EntreComp, and LifeComp), and two new frameworks created by the TWIN-IN team (Resilience Competence and Deep Tech Competence), and following a ‘learning-by-innovating approach’, Twin Transition Schools will foster skills co-creation and knowledge co-production among actors who participate in twin transitions. TWIN-IN will also produce a series of tools for responsibly enhancing twin transitions-related skill-building. https://project-twinin.eu

Along the process of identifying educational needs, UPM organized two innovative online sessions in a multi-actor approach with the aim of: 1) consolidating a consensus about “the agricultural future using digital technologies”, and 2) identifying the skills needed to make it possible, considering social, economic, environmental and technical aspects. Results from fruitful discussions among international experts on the agricultural sector using specific methodologies for prospective research and foresighting (‘Future headlines’, ‘Backwards convergency’) and online tools (Mural and IA), shaped a common view of how agriculture is evolving with the introduction of digital technologies, and defined a catalogue of skills that should be implemented into courses