What are the engineering challenges and design decisions involved in deploying a region-aware RAG system as a production web and mobile application for real farmers — and how do farmers actually use it in the field?
Most agricultural AI papers stop at "we built a model and evaluated it on a benchmark." Very few describe the full deployment pipeline — from research prototype to production system with real users. The engineering decisions (offline-first mobile, low-bandwidth optimization, AWS serverless architecture, farmer UX) are rarely documented.
This matters because the gap between "works in a notebook" and "works in a farmer's hand in a field with spotty cell service" is where most agricultural AI dies. Documenting this bridge is a contribution.
✅ Web application — deployed on AWS (Amplify + API Gateway + Lambda + OpenSearch + Bedrock)
✅ Mobile application — deployed for iOS/Android
⏳ Pilot testing with NC farmers — planned for upcoming growing season
✅ AWS serverless architecture — running in production
✅ Offline-first design for low-connectivity rural areas
⏳ Usage analytics — will be collected during pilot
AWS deployment: API Gateway → Lambda → OpenSearch (vector k-NN) → Bedrock. Why serverless. Cost analysis. Cold start mitigation. How geolocation flows through the stack.
Offline caching strategy. Low-bandwidth query optimization. Progressive loading. Touch-optimized UI for field use (gloves, sunlight, one-handed). Push notification architecture.
How we designed for non-technical users. Plain language responses. Source attribution ("Based on NC Extension AG-784"). Confidence indicators. Voice input for hands-free field use.
Usage metrics: queries per day, most-asked topics, session length, return rate. Feature adoption. Error rates. Farmer feedback (qualitative). Comparison: web vs. mobile usage patterns.
What worked, what didn't. Connectivity challenges in rural NC. Farmer trust building. Content gaps discovered through real queries. Iteration cycle from pilot feedback.
Smart Agricultural Technology (IF: 6.3) — explicitly publishes deployment papers, systems papers, and farmer-facing technology. They want to see real systems with real users, not just benchmarks. ACM COMPASS is another option — focuses on computing for underserved communities (rural farmers fit perfectly).