Govi Nena app introduced in WIDYA 

We explore the possibility of using only LLMs without incorporating any additional components such as ontologies. However, based on our literature review, we discovered that for highly technical domains such as agriculture, integrating ontologies with RAG to enhance LLMs proves to be significantly more effective. 

Status: We have drafted the system’s architecture and commenced its implementation while drafting a research paper. The expected outcomes of this research is a publication and a module to be integrated with the community section of the Govi Nena app.

Enhancing Agricultural Mobile Applications in Sri Lanka explores how mobile applications can improve yield prediction by leveraging Agent-Based Modeling (ABM) to simulate farmer interactions, crop growth, and environmental factors while addressing data scarcity through synthetic data generation. This study aims to identify essential attributes for accurate yield prediction, develop strategies to overcome data limitations, and adapt models to diverse agro-climatic zones in Sri Lanka. By integrating ABM-based predictions, Govi Nena app can provide an adaptive and scalable solution, enhancing decision-making and strengthening home gardening in Sri Lanka.   

Agriculture in Sri Lanka, particularly paddy cultivation, plays a pivotal role in the economy and in ensuring food security. However, climate change has introduced increasing levels of uncertainty in traditional farming practices. Farmers now face challenges such as irregular rainfall patterns, prolonged droughts, and shifting planting seasons, making conventional knowledge and historical trends less reliable. While research and government initiatives have attempted to address these challenges, most decision support tools currently available tend to be reactive, narrow in scope, and generalized in nature.

This research is significant in addressing several core gaps identified in the literature. Notably, there is a lack of intelligent, location aware advisory systems that simultaneously integrate climate projections and farmer specific preferences. Most existing systems in the Sri Lankan context focus on isolated aspects like pest control, fertilizer use, or irrigation scheduling. Few leverage the full potential of machine learning for predicting and recommending optimal paddy types and cultivation timelines based on forecasted climatic conditions.

The proposed system combines environmental data, weather forecasts, and farmer provided inputs to generate tailored cultivation strategies. This marks a shift from traditional decision making to a redictive, personalized model. By offering climate resilient recommendations, the research supports not only better crop management but also broader socio-economic goals, particularly for smallholder farmers who are disproportionately affected by climate risks.