93. Smartbreeder: A Data-driven Virtual Assistant to Support Nellore Sire Selection Accounting for Genotype-by-environment-by-management Interactions.

Abstract

The enviromics approach has emerged as an efficient framework for modeling genotype-by-environment-by-management (GxExM) interactions in animal breeding. It characterizes environmental conditions (ECs) in which animals are raised using detailed environmental and management data, providing a biologically meaningful representation of environmental variability for more reliable genetic evaluation and selection of better-adapted animals. Nonetheless, the implementation of enviromics in breeding programs presents two major challenges: (i) the recurrent collection of dynamic management information from producers and (ii) the translation of multiple expected progeny differences (EPDs), estimated for each sire across ECs, into practical recommendations. To address these challenges, we developed SmartBreeder, a virtual assistant that helps Nellore producers across Brazil identify sires best suited to their local climate and soil conditions, management practices, and breeding goals. SmartBreeder was developed as a web application using TypeScript, React, and Vite for the frontend, and Supabase for backend services, with PostgreSQL for data storage and Edge Functions for server-side processing. The system follows a three-step workflow: (i) collection of environmental and management information using a structured online survey, (ii) classification of the EC based on the surveyed farm profile, and (iii) recommendation of Nellore sires with their EPDs for the predicted EC. The survey includes four sections covering general information, soil management, supplemental feeding strategies, and reproductive management. A Supabase Edge Function receives twenty-two variables from the survey and queries a large language model (Gemini 2.5) via prompt engineering with few-shot learning. The prompt incorporates a training dataset of 60 farms to assign the surveyed farm profile to one of two ECs: an intensive production system (ENV1) or an extensive production system (ENV2). The predicted EC is then used to query a PostgreSQL database containing EPDs and accuracy estimates for Nellore sires in both ECs, obtained from GxExM genetic evaluations for yearling weight, scrotal circumference, age at first calving, ribeye area, fat thickness, and marbling. The platform also highlights top-ranking sires for each trait and provides complementary information (inbreeding coefficient, pedigree, and year of birth). SmartBreeder demonstrates the potential of a virtual assistant to collect high-resolution environmental and management data directly from producers and to translate multiple EPDs from GxExM genetic evaluations into practical sire recommendations through a user-friendly interface. By integrating the enviromics approach, the platform enables ECs to be assigned independently of phenotypic records, allowing sire recommendations based on GxExM models to be extended to future years and even to farms beyond the original breeding program. This approach increases the potential impact of optimized management and precision breeding across the broader cattle population. Future deployment as Android and iOS applications will enhance accessibility and on-farm adoption. SmartBreeder, developed for Nellore cattle in Brazil, can be adapted to other breeds and regions worldwide.