Running
SolarWine API
🌿
AI & ML interests
SolarWine uses AI to solve a real-time optimization problem: when should solar panels above a vineyard shade the vines versus track the sun for energy — and by exactly how much? We combine the Farquhar (FvCB) mechanistic photosynthesis model with an ML ensemble (XGBoost, Random Forest, GBR) trained on 2+ years of Negev Desert sensor data. The mechanistic model provides biological guardrails; the ML layer captures what the equations miss. A 12% divergence threshold triggers automatic safety fallback. Our roadmap includes dynamic programming for day-ahead trajectory optimization, hierarchical energy budgeting, reinforcement learning for multi-season policy improvement, transfer learning across crops and climates, and an AI advisor chatbot with anti-hallucination guardrails. Core research interests: physics-informed ML, multi-objective optimization under biological constraints, minimum-dose intervention algorithms, and responsible AI for autonomous control of physical agricultural systems.