Inspiration Agriculture remains the backbone of the global economy, yet billions in crop value are lost annually to pests and diseases that go undiagnosed until it’s too late. We were inspired by the "Plant Clinic" model used in developing nations—where experts provide community-based advice—and wanted to digitize that expertise. Our goal was to put a "virtual agronomist" in the pocket of every farmer, regardless of their location or technical background.
What it does HealthPlant is a mobile-first application that uses artificial intelligence to identify crop threats instantly.
Instant Diagnosis: Users take a photo of a distressed leaf or plant, and our computer vision model identifies pests, fungal infections, or nutrient deficiencies.
Management Strategies: Beyond identification, the app provides actionable management advice, including organic and chemical treatment recommendations.
Personalized Dashboards: It tracks the health history of specific plots, allowing farmers to monitor performance over time and receive up-to-date regional alerts about local outbreaks.
How we built it We built the platform using a modern, scalable stack:
Frontend: A cross-platform mobile app developed to be lightweight and accessible.
Backend: Hosted on Google Cloud Platform (GCP) using Cloud Run in the us-west1 region for serverless efficiency and low latency.
Machine Learning: We trained a convolutional neural network (CNN) on a vast dataset of crop-specific diseases. We utilized Large Language Models (LLMs) to transform technical pathological data into simple, conversational advice for users.
Infrastructure: We integrated the Google Cloud Monitoring API to ensure high availability and track the performance of our data science pipelines.
Challenges we ran into The primary challenge was Data Diversity. A tomato leaf with late blight looks different in the high-noon sun than it does at dawn. We had to implement robust image preprocessing to handle varying lighting conditions and low-quality cameras. Additionally, ensuring the app remained functional in areas with low connectivity required optimizing the model for on-device processing and efficient data synchronization.
Accomplishments that we're proud of We are incredibly proud of the Accuracy vs. Accessibility balance we achieved. Usually, high-accuracy models are too heavy for older smartphones, but we successfully optimized our architecture to run smoothly on a wide range of devices. We are also proud of our multi-agent AI system, which doesn't just name a disease but explains the "why" and "how" of the treatment in a way that feels human.
What we learned Building HealthPlant taught us that technology in agriculture is only as good as its usability. We learned the importance of Clean Data Science—ensuring that our datasets were "FAIR" (Findable, Accessible, Interoperable, and Reusable). We also gained deep insights into the intersection of bioinformatics and clinical-style management for non-human subjects.
What's next for HealthPlant The next phase for HealthPlant is Predictive Analytics. We aim to integrate satellite weather data and soil moisture sensors to predict disease outbreaks before they show visible symptoms. We are also looking to expand our database to include rare indigenous crops and provide a marketplace feature that connects farmers directly with sustainable treatment suppliers.
Log in or sign up for Devpost to join the conversation.