AgriDiagnose

AgriDiagnose – Detect Early, Harvest Better.

Comment

Inspiration

Agriculture remains a vital pillar of Ghana’s economy, but farmers often suffer significant yield losses due to late or inaccurate identification of crop diseases and poor access to expert advice. Many rely on visual inspection, which is prone to error. Inspired by this gap, AgriDiagnose was born—aiming to equip farmers with intelligent, accessible tools to improve their decision-making, preserve crop health, and build a more resilient farming future.

What it does

AgriDiagnose is a smart, mobile-first agriculture app designed for Ghanaian farmers. It combines cutting-edge AI with practical tools to improve productivity and sustainability. Key features include: AI-powered disease & pest detection for Maize, Cashew, Cassava, and Tomato using images taken from the farmer’s phone. Soil fertility checker based on user-inputted data for nutrient management. 24/7 AI chatbot assistant that offers expert agricultural advice on demand. Localized weather forecasts to aid irrigation, planting, and harvesting decisions. Crop calendar and planner to help farmers manage farm activities effectively. Community forum to connect farmers, share knowledge, and seek peer support. It works offline for disease detection, making it accessible even in low-connectivity areas.

How we built it

Frontend: React Native for cross-platform mobile development.

AI Models: Deep learning using TensorFlow and PyTorch.

EfficientNetB3/B4 (Cashew, Cassava, Maize)

MobileNetV2 (Tomato – optimized for mobile)

Model Training:

Transfer learning with progressive unfreezing

Extensive data augmentation for real-world variability

Mixed precision training for performance

Backend Tools: Firebase (authentication, storage), REST APIs (weather), SQLite (offline data handling)

Chatbot: Built using a fine-tuned language model integrated via a lightweight NLP engine.

Challenges we ran into

Lack of localized, labeled datasets, especially for diseases specific to Ghana’s crops.

Optimizing AI models to work efficiently on low-end smartphones with limited memory.

Ensuring real-time performance without sacrificing accuracy.

Designing an intuitive UI for farmers with varying levels of literacy and tech experience.

Seamless integration of multiple components: AI, weather, calendar, chatbot, and forum.

Accomplishments that we're proud of

Developed four high-accuracy AI models, each tailored for a major crop.

Achieved over 90%+ test accuracy for all disease classifiers.

Created a fully functional, offline-capable mobile app ready for real-world use.

Built a system that combines practical farming tools and advanced ML in a way that is usable by farmers.

Received positive feedback during pilot testing with local farmers and agricultural officers.

What we learned

Real-world applications of AI require balancing technical performance with usability.

Building for offline use forces thoughtful design and optimization choices.

Even the best models are useless if the user interface isn't intuitive—farmer-first design is key.

There’s immense value in co-creating with end users, not just building for them.

What's next for AgriDiagnose

🚜 Pilot deployments in key farming regions across Ghana.

🧪 Continuous model improvement through real-world image feedback and retraining.

🗣️ Multilingual support (Twi, Ewe, Dagbani) to increase accessibility.

📈 Expand crop support (e.g., Cocoa, Yam, Pepper).

🌐 Offline chatbot capabilities and localized treatment recommendations.

🤝 Partnerships with NGOs, government extension services, and agricultural input suppliers.

🛰️ Future expansion to include satellite/droned imagery and IoT sensor integration.

Built With

Share this project:

Updates