TOSCAN
Specialised AI-based advice for small farmers
Challenge
Small-scale agriculture currently plays an important role in the resilience, sustainability and security of the agri-food sector. However, this type of agriculture continues to be the most affected by challenges associated both with climate change and the lack of specialised information for its professionals. Although the widespread use of the internet has provided these farmers with a range of additional tools, the quality of the information available is still a concern. In fact, the use of incorrect information jeopardises not only the productivity and competitiveness of the farm, but can also result in the implementation of practices that are unsafe from a food and environmental point of view.
Solution
The TOSCAN project proposes an intelligent advisor to help farmers make decisions on issues such as crop diseases and pests, fertilisation and irrigation, taking advantage of the multidisciplinary nature of the consortium involved. The project aims to go beyond current chatbot systems in 4 key factors: 1. multimodal interaction with the farmer (via text, voice and/or images); 2. personalisation of information based on specific data on crops, soil, atmosphere; 3. reliability and security of the information provided; 4. ability to interpret regionalisms, accents and jargon to facilitate communication between farmers and the digital advisor.
Objectives, Activities and Results expected / achieved
Specific expected objectives:
O1: To develop and implement a functional chat model, TOSCAN LLM, capable of answering specific questions, in a personalised way, on sustainable agricultural practices, particularly with regard to crop protection, nutrition and irrigation;
O2: Promote the implementation of more sustainable agricultural practices by small-scale farmers;
O3: Develop and implement an effective security model (guardrails) that guarantees the security of the information and suggestions provided by the system, identifying and mitigating risks associated with potentially dangerous or inappropriate suggestions, by the end of the second year of the project;
O4: Develop a speech-to-text model capable of correctly transcribing the different accents, regionalisms and linguistic variations within the regions covered during the project;
O5: Identify diseases and pests in apple, pear and grapevine trees based on photographic images of visible symptoms.
Project Reference
CENTRO2030-FEDER-01183900Funding
Intervention Region
PortugalTotal Investment
523.419,20IPN Investment
249.078,40Total Eligible
523.419,20IPN Eligible
249.078,40EC Funding – Total
413.199,72EC Funding – IPN
211.716,64Duration
36 MonthsStart Date
2025-01-01End Date
2027-12-31Approval Date
2024-10-15Consortium
Impactwave, LdaInstituto Pedro Nunes
Universidade de Coimbra
Keywords
Agriculture;LLMs;
Artificial Inteligence.
