FarmHand

Inspiration

We started with a simple observation: farmers are losing 20 billion pounds of produce every year because they can't accurately assess their soil and crops. Soil degradation is happening at an alarming rate with 100 million hectares lost per year. Most farmers don't have good tools to monitor or predict what's happening in their fields. The problem felt fixable. Farmers needed real-time insights, but most soil testing requires sending samples to labs. We thought: what if we could give them instant feedback right in the field?

What it does

Farmhand uses Rayneo AR glasses to help farmers assess soil health on the spot. Here's how it works: Soil Analysis: Take a photo of your soil through AR glasses. Our system analyzes texture, crusting, erosion, and drainage. Crop Analysis: We pull in soil data and geospatial information to predict how crops will perform. Growth Simulation: See how your field might look over time based on current conditions. The whole thing can be voice-controlled as well, which matters because farmers have their hands full in the field.

How we built it

We used GPT-4o to process soil images and extract useful data. We built on Cornell's CASH framework for soil assessment. We chose RayNeo AR glasses because they're lightweight, work with voice commands, and connect via ADB.

Challenges we ran into

Camera: Enabling the camera is harder than it sound. Building trust: Farmers have years of experience reading their land. We needed our recommendations to make sense to them, not just come from a black box. Real-time performance: Running vision models on AR hardware while keeping everything smooth and responsive required optimization.

What we learned

The math worked out. On a typical farm with moderate costs and reasonable yield improvements, farmers break even in under one season. That's the difference between an interesting project and something people will actually use.

We also learned that soil health isn't abstract. Farmers care because it directly affects their bottom line.