Agrinix

Snap. Scan. Save your crop.

Comment

Inspiration

Growing up with my grandma and uncles in an agricultural community, I realized and heard stories of my grandma and uncles complaining of pests destroying their crops, which majorly included cocoa. This was because my parents could not early-detect the crop diseases and wait for them to cause damage or spread out before the damage is realized. It is for this reason I thought it wise to come up with an app that will allow farmers, not only my parents, easily detect diseases that affect their crops to alleviate the loss that comes with this "ignorance". This solutions is;

  • Fast
  • *Accessible *
  • Easy to use

What it does

The Agrinix app is an app that allows farmers to capture images of leaves of infected crops, the application analyses the crop and provides a response with predictions, symptoms, causes and possible cure. So in general, the app;

  • Takes an image of an infected crop
  • Analyze the image
  • Provide disease predictions
  • ** Provide possible cure for the disease detected.
  • Community session for farmers to seek support from fellow farmers or to provide support to fellow farmers.
  • Provide accurate climate questions for the farmer to guide in farming practices
  • Provide knowledge to farmers on possible crop infections and how to cure them.
  • Store detection history for future access.

How we built it

The project consists of the following components:

1. Frontend (Flutter Mobile App)

  • Built with Flutter for cross-platform support (Android-first).
  • Simple UI to capture for easy usage.
  • High optimization and scalability.

- I used highly optimized flutter packages such as image_picker, for image capture and picking images from gallery, share_plus for sharing contents of the app to third-party platforms like whatsapp.

2. Machine Learning Model

  • Trained a model to detect the crop diseases using Roboflow and the image classification model.

- Dataset: Publicly available plant disease datasets (PlantVillage, etc.)

3. Model Conversion & Integration

  • Trained the model with the Roboflow image classification model.
  • Deployed the model of Roboflow workflow.
  • Made API requests to the Roboflow deployment endpoint to get prediction results.

- Passed the Roboflow prediction results to deepseek to get symptoms, causes, cure etc. of the disease class.

4. Optional Backend (for model updates & logging)

  • Nest.js was used for developing the server side of the application, using typescript.
  • Images received from the flutter app were processes with cloudinary.
  • The image's url string is passed to the Roboflow model for detection.
  • The returned Roboflow response's disease class is sent to deepseek via an API call to get the symptoms, causes, cure, etc. for the particular disease.
  • The returned results from deepseek are parsed and sent to the frontend flutter app.

- The results are stored in a database form persistence.

Challenges we ran into

  • Dataset imbalance: Some diseases had far more samples than others. I applied data augmentation techniques to reduce bias.
  • Model size vs. accuracy: Had to balance model complexity with mobile constraints — especially for older Android phones.
  • Image preprocessing: Lighting and background noise in images affected prediction accuracy. Added normalization and cropping.
  • Time constraints: Training and testing multiple models in the limited hackathon timeframe was intense!

Accomplishments that we're proud of

  • We are very proud to have completed the MVP of this project, although not fully completed,, we are glad we built something that works for everyone to use.

What we learned

  • How to train and optimize ML models for edge devices.
  • Deployment of ML models.
  • Real-world challenges in building human-friendly apps.
  • UX design principles for low-tech literacy users (e.g., farmers).

We also gained deeper appreciation for the impact of AI in agriculture and the technical skills needed to bring that vision to life.

What's next for Agrinix

  • Add support for more crops and diseases.
  • Improve accuracy with better datasets and fine-tuned models.
  • Add multi-language support for local farmers.
  • Explore integration with SMS alerts or WhatsApp chatbots.
  • Enhance community support to include live chat sessions for farmers.
  • A market forum for farmers to display their crops to potential buyers and facilitate easy marketing and selling of farm products.

Quick Links

Frontend Github repo Backend Github repo Frontend app documentation Backend server documentation

Built With

Share this project:

Updates