AgriDoc AI

Inspiration

Farmers lose 30–40% of yield annually due to undiagnosed plant diseases. Existing apps are either limited in crops, or not accessible offline, or require technical knowledge. We wanted to build an Agentic AI system that works like a digital farm doctor — fast, simple, and accurate, even for layman farmers.

What it does

Farmers upload a leaf image (mobile/web for time being only web). AI detects the crop type (Corn, Rice, Potato + 38 more crops). Identifies the disease (e.g., Rice Blast, Corn Leaf Blight). Gives remedy suggestions: pesticide usage, natural treatments, and best farming practices.

Works in low-bandwidth rural settings using lightweight models.

How we built it

Dual AI model approach: 1)Wide TensorFlow model (.h5) → recognizes 38 crops + diseases. 2)Specialized PyTorch model (.pth) → trained on 30GB dataset for Corn, Rice, Potato. Designed an intelligent router agent: If crop = Corn/Rice/Potato → uses specialized model (high accuracy). Else → falls back to wide model. Agent uses tool which is ML model Built in Python + TensorFlow + PyTorch. UI prototype: For time being Ui is simple streamlit application

Challenges we ran into

Handling huge dataset size (30GB) → required efficient preprocessing. Model conflicts (wide model predicting “Strawberry” for Rice Blast!) → solved via dual-model routing. Balancing accuracy vs. speed for real-time predictions. Packaging models so they can run on friends’ systems without dataset.

Accomplishments that we're proud of

Achieved >90% accuracy on Corn, Rice, and Potato with specialized models. Built a scalable architecture → easy to add new crop models. Created a system that even a layman farmer can use → just upload a picture. Hackathon-ready deployment (runs locally, shareable .h5 + .pth models).

What we learned

Importance of model specialization instead of “one-size-fits-all”. Integration tricks for TensorFlow + PyTorch models in one pipeline. Building explainable outputs for non-technical users. The power of agentic AI in agriculture.

What's next for AgriDoc AI

Better End-End System Add more crops & regional diseases. Integration with farmer advisory services (open source alerts model). Use satellite + drone imagery for large-scale farm monitoring. Build B2B partnerships with AgriTech firms and government bodies.

Built With

• deepseekr1
• llm
• natural-language-processing
• python
• pytorch
• resnet
• streamlit
• tensorflow