Inspiration
- In regions with limited access to advanced equipment or agronomical support, smallholder coffee farmers often rely on guesswork to anticipate yields. YieldBrew was inspired by the idea of using lightweight AI tools to simulate and support farming decisions without requiring expensive hardware or large datasets.
What it does
- YieldBrew is an AI-powered tool that predicts coffee yield (kg/ha) based on various environmental and agricultural input features like rainfall, temperature, soil quality, and fertilizer use. It uses a simple regression neural network and can run locally on low-cost devices. The tool is demonstrated using a Gradio interface that allows users to input custom values and get real-time predictions.
How we built it
- Model: A PyTorch-based regression neural network (ReNN) trained on synthetically generated coffee farming data.
Data: Synthetic data was created to simulate realistic agricultural conditions and yield outcomes.
Demo Interface: Gradio was used to build a user-friendly interface for entering conditions and displaying predictions.
Challenges we ran into
Data availability: Real-world coffee yield datasets with environmental features were extremely limited or inconsistent. We had to simulate our own data carefully to make model training feasible.
Practicality: Building a useful model on synthetic data was a balancing act — it needed to be realistic enough to behave sensibly, even if not field-validated.
Import and packaging issues: Getting Python module imports to work seamlessly across different folders and environments took some debugging, especially for demo deployment.
Accomplishments that we're proud of
Built a functional end-to-end AI tool in under a week as a solo project.
Created a realistic synthetic dataset that allowed meaningful yield prediction behavior.
Packaged everything into a clean, modular repository with a demo interface.
Log in or sign up for Devpost to join the conversation.