CropGuard_Agentic

Inspiration

What Inspired Us

Smallholder farmers across Africa face severe crop loss due to plant diseases and nutrient deficiencies, often lacking immediate access to expert agronomists. We built CropGuard Agentic AI to bridge this gap by bringing dynamic, localized crop diagnostics directly to the field—even in bandwidth-constrained or completely offline environments.

How We Built It

CropGuard relies on a lightweight, quantized autonomous AI model (CropGuard-Agronomist-1.1B-Q4_K_M) powered by llama.cpp for low-resource deployment.

The core architecture follows an agentic orchestration workflow:

1. Vision Confidence Routing: When a farmer scans an infected crop leaf, the system evaluates the visual confidence score ($C$). If $C < 70\%$, the request automatically reroutes to an offline local Retrieval-Augmented Generation (RAG) agronomy knowledge base.
2. Multi-Agent Decision Pipeline: An agronomy diagnostic agent evaluates the nutrient profiles (such as nitrogen or phosphorus deficiencies) and cross-references them with regional language needs.
3. Localized Action Plan: The localized output agent translates pathology metrics into actionable organic field remediation plans across English, Swahili, Hausa, Yoruba, and Igbo.

What We Learned

Deploying large language models on budget hardware requires aggressive quantization ($GGUF \text{ Q4_K_M}$) and specialized system design. Balancing edge runtime limits with high diagnostic accuracy taught us how crucial load-bearing cross-disciplinary pairing is between AI architecture and real-world agricultural science.

Challenges We Faced

• Hardware & Bandwidth Limits: Optimizing dynamic multi-agent decision paths to execute on budget devices under strict memory overhead.
• Multilingual Agronomy: Ensuring accurate translation of technical plant pathology terms across local African languages without losing clinical intervention context. ## What it does

How we built it

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for CropGuard_Agentic

Built With

• agriculture
• autonomous-agents
• django
• docker
• edge-ai
• gguf
• langchain
• langgraph
• llama-cpp
• postgresql
• python
• quantization
• rag
• react