RISK MANAGEMENT TOOL TO INCREASE FARM SAFETY

VISIMO proposes the development of a customizable decision-support tool, Demeter, which will allow producers to assess and mitigate risk in real time, increasing the safety, efficiency, and productivity of small and mid-size farms.The primary outcome of the Phase II effort will be a prototype of the Demeter application. Demeter will accurately assess risk and provide mitigation suggestions, including cost-benefit analyses, for each suggestion. As development progresses, improvements will be made to the GNN and to the arithmetic formula to improve risk mitigation. Over the period of performance, VISIMO will collect between 5,000 and 7,000 data entries on small and mid-size beef and dairy farms.Phase II Technical Objectives:Gather Requirements:Formally document R&D specifications for the work identifying necessary features. Establish metrics and benchmarks for future validation. Recruit testers an d end users.Enhance PWA and Backend:Gather preferences from potential end users and design a user-friendly User Interface (UI). Test functionality. Iterate the online backend, using Django with GraphQL; use React JS for the frontend.Develop NLP and Object Detection Tools:Add features that incorporate data uploaded by users and that uses NLP to facilitate data. These will be established early for ongoing data collection. Models will be trained on real user data to support NLP and CV.Conduct Alpha Testing:Identify end users and deliver an Alpha version of the PWA to testers; assist with install and use. Collect user feedback, focusing on functionality issues and support, continually gathering data and feedback to iteratively improve components of Demeter. Compile data points collected in Alpha testing for use in updating synthetic data and train the GNN.Conduct Beta Testing:Deliver Beta version of PWA to testers and assist with install, use, and support. Collect feedback from users, focusing on usability concerns. Analyze usability concerns and enhancement prioritization, gathering feedback to iteratively improve components of the application. Continue compiling data points collected during Alpha testing, updating synthetic data for GNN training.Perform Iterative Improvements:Release a 1.0 version of the PWA available to a general audience, allowing users to provide feedback. Review Phase I assumptions in synthetic data generation against incoming real data and make corrections. Use data to continue improving the existing ML components of the application.Proposed Phase II Success Metrics:Process Based:Triplet Loss: Measures how well embedding space translates similarities and differences between hazards.Data Quantity: Number of data entries collected.Application Performance: Metrics tracking speed, bug frequency, and prominence.Usage Statistics: Quantity and demographics of users.Outcome Based:Precision: Fraction of actual accidents among predicted accidents.Recall: Fraction of actual accidents among predicted non-accidents.User Satisfaction: Survey results about user experience.MAE: The amount of agreement between the arithmetic formula and SMEs.